Theoretical Calculus by Clark 1972 Lite

The Theoretical Side of Calculus
The Theoretical Side of Calculus
Colin Clark
University of British Columbia
Robert E. Krieger Publishing Company

Huntington , New York
1978
Original Edition 1972
Reprint 1978 with corrections
Printed and Published by

ROBERT E. KRIEGER PUBLISHING CO., INC.
645 NEW YORK A VENUE ~
HUNTINGTON, NEW YORK 11743
Copyright 1972 by
WADSWORTH PUBLISHING COMPANY, INC.
Transferred to Colin Clark, 1977
Reprinted by Arrangement
All rights reserved. No reproduction in any form of this

book, in whole or in part (except for brief quotation in
critical articles or reviews), may be made without written
authorization from the publisher.
Printed in the United States of America
Library of Congress Cataloging in Publication Data
Clark, Colin Whitcomb, 1931-

The theoretical side of calculus.
Reprint of the edition published by Wadsworth Pub. Co.,
Belmont, Calif., with corrections.
Includes index.
1. Calculus. I. Title.
[QA303.C58 1978] 515 78-6731
ISBN 0-88275-680-X
Preface
A standard feature of the present mathematics curriculum at many

universities and colleges is the separation of elementary calculus courses into
two types, "regular calculus" and "honors calculus." Unfortunately, there are
some serious disadvantages, primarily of a psychological nature, inherent in
such a division. The "regular" students may resent what they consider to be
"inferior" treatment and consequently learn less than they would otherwise.
Instructors prefer to teach the "honors" classes, so that the less-experienced
instructors may be assigned to "regular" calculus courses, a process that further
degrades the "regular" classes.
Moreover, the "honors" classes do not always turn out so well as expected.
The mixing of practical and theoretical material throughout the course results
in considerable confusion. Mastery of both aspects turns out to be more difficult
than many of the students had anticipated. Since transferring back to the regular
course involves an admission of defeat, some students of good ability give up
mathematics in disillusionment.
This text is designed to offer an alternative approach, in which the theoreti-
cal part of elementary calculus is studied after the technical aspects of the
subject have been mastered in an earlier course. Students of outstanding ability
who are preparing for research careers in mathematics will be ready for this
course at the sophomore (or perhaps even the freshman) level. It may also
interest others-for example, prospective mathematics teachers-who wish
to deepen their understanding of calculus beyond the formalities of differentia-
tion and integration. It can also be used to provide a transition to more ad-
vanced areas of mathematics, such as multivariate calculus, abstract analysis,
and complex function theory.
Preface
This text introduces students to mathematical analysis at as elementary a

level as possible and proceeds far enough to develop rigorously the theorems of
basic one-variable calculus. Prospective r eaders are assumed to be familiar
with the techniques of differentiation and integration of functions of one vari-
able. Our logical starting point is the real-number system, the properties of
which are taken as axioms.
The first two of the five chapters are especially elementary, covering the
concepts of limit and continuity and the properties of the real-number system.
Limits are first discussed for the case of sequences; the transition to limits of
functions (Chapter 2) is then fairly easy. Chapter 3 is devoted to proving the
basic theorems about continuous functions. Calculus appears for the first time
in Chapter 4, where the following topics are treated rigorously : differentiation
and integration, convergence of power series, and transcendental functions.
Chapter 5 gives a brief introduction to limits and continuity in n dimensions and
concludes with two famous examples which indicate that continuity is not so
simple a concept as one might expect.
Of the two appendices, the second, on mathematical induction, is a pre-
requisite to the text proper, and should be studied by any student not familiar
with this topic. Appendix I, on logic, should be studied whenever the need
arises.
If taken at a leisurely pace of, say, two class hours per week, the entire text
could be covered in two semesters. Most of Chapters 1-4 could probably be
given in three hours per week for one semester. Almost everyone finds his first
steps in rigorous analysis rather difficult, however, and I strongly recommend
the slower pace where feasible. .
It is a pleasure to acknowledge my indebtedness to many colleagues at the
University of British Columbia. Several who taught classes from a preliminary
version of this book contributed from their experience. The influence of Klaus
Hoechsmann, Ron Riddell, and Maurice Sion was particularly strong. The
manuscript was beautifully typed against tremendous odds, by Miss Barbara
K.ilbray.
Vancouver, B.C.
September, 1971
Contents
)
Chapter 1 Sequences, Limits, and Real Numbers
1.1 Sequences 1
1.2 The Method of Iteration 3
1.3 Properties of the Real Number System 5
1.4 The Limit of a Sequence (Introduction) 9
1.5 The Limit of a Sequence (Numerical Examples) 13
1.6 The Limit of a Sequence 16
1. 7 Elementary Theory of Limits 21
1.8 The Completeness Property 27
1.9 The Base of Natural Logarithms 33
1.10 Growth Properties of Certain Sequences 37
1.11 Some Further Properties of the Real Number System 40
1.12 Infinite Series 44

Chapter 2 Limits, Continuity, and pifferentiability
2.1 Introduction 54
2.2 The Limit of a Function 59
2.3 Other Types of Limits 66
2.4 Continuity 71
2.5 Differentiability 81
Chapter3 Properties of Continuous Functions
3.1 Introduction 91
3.2 Supremum and Infimum 92
3.3 The Bolzano-Weierstrass Property 98
3.4 Cauchy Sequences 101
3.5 Properties of Continuous Functions 103
3.6 Uniform Continuity 107
Chapter 4 Some Theorems of Calculus
4.1 The Mean Value Theorem 111
4.2 The Riemann Integral 115

4.3 Integrals and Derivatives 129
4.4 Improper Integrals 133
4.5 Uniform Convergence 138
4.6 Power Series 147
4.7 Uniform Convergence of Power Series 152
4.8 Taylor's Theorem 159
4.9 On the Definition of the Exponential Function 163
ChapterS Limits and Continuity in n Dimensions
5.1 Open and Closed Sets in Bin 169
5.2 Sequences and Limits 176
5.3 Limits and Continuity 180
5.5 Two Famous Examples 193
Appendix I Logic
1.1 Logical Connectives 201
1.2 Quantifiers 206
1.3 Proof 210
1.4 Sets and Functions 212

Appendix II Mathematical Induction
II.l Mathematical Induction 216
II.2 Mathematical Induction (continued) 222
11.3 The Binominal Theorem 224
Solutions to Selected Exercises 229

Index 237
Index of .Special Symbols 240
Historical Introduction
The Greek scholars were the first to consider seriously problems of con-.
tinuity and infinity. These problems depend in turn on the basic concept of
"number." Being excellent geometers, the Greeks attempted to include the
concept of number in their geometry. Every line segment had a length (relative
to a given unit length), which was expressed by a number; conversely every
number was thought to be the length of some line segment. In this way, opera-
tions between numbers could be realized by means of geometrical "ruler and
compass" constructions.
The construction of a line segment of rational length mfn (where m, n are
integers, n =!= 0) is an easy matter. The hope that all numbers were rational was
destroyed by the discovery, around 400 B.c., that .J2 is irrational, even though it
corresponds to a constructible line segment, the diagonal of a unit square. It is
said that the Pythagorean mathematicians were so embarrased by the irration-
ality of .J2 that they "classified" this infor~ation, and the first scholar to "leak"
it to the public was poorly treated by his colleagues.
Three famous classical problems, those of"trisecting the angle," "doubling
the cube," and "squaring the circle," were closely related to the question of
realizing all numbers as lengths of constructible line segments. In modern terms
these problems are equivalent to constructing, by ruler and compass, line
segments of length cos 20 (for example), ~. and 1r, respectively. In the
nineteenth century all three constructions were shown to be impossible.t Thus
t P. L. Wantzel (1814-1848) proved thilt cos 20 and t'2are nonconstructible. C. L. F.

Lindemann (1852-1939) proved that '"is "transcendental," meaning that'" is not the root of any
algebraic equation with integer coeffici~nts, and hence is also nonconstructible.
xi
xii Historical Introduction
the Greeks' concept of number was destroyed completely, and mathematicians

were still faced with the task of defining number unambiguously and independent
of geometrical considerations.
Toward the end of the nineteenth century several successful theories of
the real-number system were developed. The best known are those of Karl
Weierstrass (1815-1897) and J. W. R. Dedekind (1831-1916). Both defined real
numbers in terms of infinite sets of rational numbers. Later Georg Cantor
(1845-1918) showed that the real numbers form an "uncountable" set, in the
sense that, unlike the rational numbers, they cannot be placed in one-to-one
correspondence with the set of positive integers. From this it follows that
some (in fact most) real numbers cannot be obtained by any finite process from
the rational numbers.
The work of Weierstrass and Dedekind showed that the real-number systeni
possesses a fundamental property not possessed by any smaller system (such
as the rationals). This property, called completeness, can be described as follows:
every nondecreasing, bounded sequence of real numbers "converges" to a real
number. It follows from the completeness property that every decimal
terminating or not, represents a real number, and conversely. This shows how
to locate any real number, at least to any desired degree of accuracy, as a: point
on the "real line." Thus the Greek quest of identifying numbers with lil\,.e
segments was realized at last. Only the desire to obtain every number by (finite)
construction turned out to be unrealizable.
Questions of continuity and infinity seem to have represented a complete
mystery to the Greek scholars. The difficulties were clearly indicated by the
famous paradoxes of Zeno (ca. 450 B.c.), of which we quote the following:
(a) The paradox of Achilles: In a race between Achilles and. a tortoise, the
latter has been given a head start. In order to catch up with the tortoise,
'Achilles must first cover half the distance separating them. Then he
must cover half the remaining distance and so on. Hence, at any
time, Achilles still has at least half the distance to go, and consequently
is never able to pass the tortoise.
(b) The paradox of the arrow: Consider an arrow in flight. At any instant,
the space occupied by the arrow is equal to the length of the arrow.
Hence the arrow cannot undergo motion at any instant. Since time is
composed of instants, no motion of the arrow is possible.
The paradox of Achilles seems to indicate that a continuous model of
physical motion is self-contradictory, whereas the paradox of the arrow seems
to show that a discrete model is equally self-contradictory. Speculation on
matters of this sort later became a trademark of medieval theologism.
Historical Introduction xiii
The modern era in mathematics and physics begins with one of the greatest
geniuses of all time, Isaac Newton (1642-1727). In a period ofless than two years,
1665-1666, there occurred the most cataclysmic intellectual revolution of all
history. Working alone at his country home in order to escape the plague,
Newton
(i) invented the differential and integral calculus, based on a study of
infinite series;
(ii) established the basic laws of motion and of gravitation;
(iii) using the results of (i) and (ii), solved completely the ancient problem
of the motion of the planets; and
(iv) discovered the nature of light.
Except for the work on light, Newton's discoveries went unpublished for over
20 years. Meanwhile G. Leibniz (1646-1716) discovered the calculus independ-
ently and published his theory. A stupid argument over priority arose between
the followers of Newton and those of Leibniz. That Leibniz himself recognized
Newton's accomplishment is obvious from his remark, "taking mathematics
from the beginning of the world until the time of Newton, what he has done is
much the better half."
In spite of the great success of their theories, neither Newton nor Leibniz
was able to give a convincing explanation of the logical principles underlying
the calculus. Both men struggled in vain with the concept of "infinitesimal";
Leibniz, in particular, committed several errors in questions relating to in-
finitesimals. The ensuing two centuries saw rapid and far-reaching developments
in science and mathematics, and among practical-minded men the feeling grew
that logical speculations were best avoided. In view of the tremendous success
with which calculus could be applied to numerous physical phenomena, this
viewpoint was clearly justified.
It is perhaps surprising, therefore, that any mathematicians remained
sufficiently stubborn to insist on trying to understand the basic principles of
their subject. But by the early nineteenth century certain developments in
mathematical physics forced mathematicians to consider once more the basic
problems of number, continuity, and infinity. During his work on the theory of
heat, J. B. J. Fourier (1768-1830) emphasized the importance of trigonometric
series,
b0 +a 1 sin x +b
1 cos x +a 2 sin 2x +b 2 cos 2x + .
It soon became evident that such series behave quite differently from "pgwer
series,"
which were extensively used by Newton and his successors. We know now that
if a power series converges, then the sum always represents a continuous (in
fact, infinitely smooth) function . Fourier himself realized, on the other hand,
xiv Hlatorlcallntroductlon
that a trigonometric series could converge to a discontinuous function such as

those shown below. Some mathematicians criticized Fourier's work by asserting
r--"'1
I I
I I
I I I
I I I
L-...J L....
that such graphs do not represent actual "functions," that Fourier's concept of
"convergence" was faulty, and so on. To settle these controversies, unambiguous
definitions of such concepts as "function," "convergence," and "continuity"
were needed. This brought mathematicians back to the problems of Zeno
and to the necessity of developing calculus on a rigorous logical basis.
Among the earliest and most significant contributors to rigor in calculus
was A. Cauchy (1789-1857). In his text Le~ons sur le calcul differentiel (1829),
Cauchy defined the derivative as the limit of a quotient:
dy = limf(x +h)- f(x).

dx h~o h
Cauchy explained the meaning of limit in the following terms:
When the successive values attributed to a variable approach in-

definitely a fixed value so as to end by differing from it by as little
as one wishes, this last is called the limit of the others.
Even this definition seems excessively vague by modern standards: the phrases
"successive values," "variable," "approach indefinitely," "as little as one
wishes," are all suggestive rather than precise.
Finally in 1872, H. E. Heine (1821-1881) presented the following formula-
tion of the definition of the limit of a function f (x) at x 0 :
"If, given any e, there is an fJo such that for 0 < 1J < fJo the differ-
ence f(x 0 1}) - L is less in absolute value than e, then L is the
limit ofj(x) for x = x 0 ."
This statement, which is now the accepted definition of limit, is absolutely un-
ambiguous. With minor modifications, it applies to many other kinds of
limiting processes, including sequences and series of numbers and functions,
functions of several variables, complex functions, and so on. The paradoxes of
Zeno regarding time and motion disappear once the definition of continuity
based on Heine's definition of limit is understood. There is probably no other
Historical Introduction xv
instance in human intellectual history in which so much time and effort was
spent merely to reach a satisfactory definition!
Given clear definitions of number, limit, continuity, and derivative,
nineteenth-century mathematicians were able to provide a logically precise
development of the calculus. The trigonometric series of Fourier became
acceptable, thanks to studies by P. G. L. Dirichlet (1805-1859) and others. Of
particular significance was the proof, by Weierstrass and B. Bolzano ( 1781-
1848), that any continuous function f(x) defined on a finite closed interval
a ::::;: x s b must assume finite extreme values.t It is clearly important to be able
to deduce such "obvious" facts from given axioms and definitions.
In the twentieth century great developments in both pure and applied
mathematics have been built on the foundations laid in the previous century.
Although these developments are; beyond the scope of this book, let us mention
one particularly interesting example, the study of spaces of infinite dimension,
initiated primarily by D. Hilbert (1862-1943). These spaces, whose properties
obviously cannot be deduced by "geometrical" intuition, play an important role
in modern physics, especially quantum mechanics.
By and large, the mathematics of this century is characterized by an un-
limited capacity for generalization and abstraction. As in previous eras, much
contemporary mathematical research seems to have little to do with practical
affairs. Yet modern civilization is experiencing a rapidly increasing level of
"mathematization," accelerated especially by the development of electronic
computers. The major technical problems of the future (including the critical
problems of environmental distress) are certain to require highly sophisticated
modern mathematical techniques. People who clearly understand both the
potentialities and the limitations of mathematical models will continue to be as
rare as they are valuable.
(For further information on the history of calculus, see Carl B. Boyer,
A History of Mathematics, Wiley (1968).)
t It should be pointed out that a certain amount of controversy is still attached to the
meaning of theorems such as this, on the grounds that no constructive methods are possible for
determining extreme points in general. See Appendix I.
J Sequences, Limits, and Real Numbers
1.1 Sequences
The intuitive idea of an infinite sequence,
Xl, X2, Xs, ' Xn, ' (1.1)

t ~ t t
I 1 1 " " -,
can easily be given precise mathematical meaning.
Definition A sequence Q:J is a function defined for all positive integers

n = l, 2, 3, . . . . Instead of the function notation x(n), we use the subscript
notation ~for the nth term of the sequence.
An example is the sequence {nf(n + 1)}, in which Xn = nf(n + I), or

written in detail,
-I -2 -3 n
2'3'4'"""'n+1'
Sequences are frequently specified by simply giving a formula for the nth
term. For example if Xn = ( -l)n, the sequence {xn} is
-I, I, -I, ... , ( -1 )n, ....
Another useful method for specifying a sequence is by means of a
1
1 Chapter One Sequences, Limits, and Real Numbers
recursion formula. As a simple example, consider the sequence
1, 3, 6, 10, ?, ?, ....
(Such "sequences" used to occur on intelligence tests.) The terms given satisfy
the following conditions:
x 1 = 1,
x,.+l = x,. + (n + 1) (n ;;::: 1).
Any formula (such as the one just above) of the form
x,.+l = .(3x..) (1.2)

d, r;,')'~jl rL o-( l1J'J1l/l)oi-J{
is called a (two-term) recursion formula for the sequence {x8 }. It is sometimes
possible to derive an explicit formula for X 8 from such a recursion formula.
Thus, in the example xn+1 = X 8 + (n + 1), we have:
x 1 = 1,
x2 = 1 + 2,
x3 = 1 + 2 + 3,
X8 = 1+2 + 3 + + n
= in(n + 1).
(On the last line we have used a well-known formula for 1 + 2 + 3 + + n;

see Appendix II.)
Exercises
1. Write out the first few terms of the sequences.
(a) {2n}, (b) {cos n1r},
(c) {(-1); + 1} (d) {In - 31 - 3}.

1.2 The Method of Iteration 3
2. Find simple sequences {x,.} which begin as shown.

(a) I, -t, l, -!, ... , (b) 1,3,5,7, ... ,
(c) I,O,I,O, ... , (d) 2, 5, 10, I7, ... '
(e) 4, 6, 10, 18, ... , (f) 1, 2, 6, 24, ... .
3. Solve the following recursion formulas-that is, find x,. in general. Assume
xl = 1.
(a) x,.+1 = x,. + 2, (b) x,.+l = 1- x,.,
n+1
(c) x,.+l = x 1 + x 2 + + x,., (d) Xn+l = - - x,.,
n
(e) x2 = -1 and x,+a = x,..

4. Given the information
x1 =I; x,+l = x,. + n + I (n :;;::: 1),
prove by mathematical induction that x,. = tn(n + I) for all n.

5. Use mathematical induction to prove that your answers to Exercise 3 are
correct.
1.2 The Method of Iteration
Consider the problem of determining the value of .J2

to a given degree of
accuracy. If x = .J2, then of course x 2 = 2. By simple algebra we can rewrite
this in the strange form
x=x - -2.2
+ (1.3)
2x
Equation (1.3) is of the form
x =f(x). (1.4)
The method of iteration for solving such an equation consists of constructing

a sequence by the recursion formula
x,+l = f(x,.). (1.5)

Also x 1 has to be chosen to begin with. For Equation (1.3) we proceed as follows.
Let x 1 = 1 (a first "wild guess" for .J2). Then using

X~+ 2
Xn+l =-2--, (1.6)
x ..
we get
1+2
X2=-- = 1.5,
2
(1.5) 2 + 2
= 1.417,
Xa = 2(1.5)
X
4
= (1.417)2 + 2 = 1.4142
2(1.417) '
and so on.
It is a fact (which I hope you find amazing at this point) that the sequence
{xn} constructed in this manner "converges" to .J2 as n becomes large. It can
even be shown that Xn+l is a better approximation to .J2 than xn is, by at least
two decimal places. We will return to this question later. At the present time
let us prove that ~, ;
Xn ?:. .J2 (for n ?:. ,2): >- (1.7)
!
To see this, we use the inequality a + 2 b2

?:. 2ab (why is this inequality true?).
Thus ~~ \ ~f; ((A-~)-,__/ o
Xn+l =X~+ (.j2)2 ?:_ 2.fi Xn = .J2,
2xn 2xn
for n ?:. 1, which proves (1.7).

The sequence {xn} of (1.6) will be studied in greater detail in Section 1.8.
Exercises
1. Find a recursion formula for calculating .J~ (a > 0). Try it on .Ji
~~
Let {xn} be the sequence of Equation (1.6). Using (1.7), show that {x..}
(?) is a nonincreasing sequence, in the sense xn+l :s:; Xn (for n ?:. 2).
3. Let {xn} be the above sequence. Prove that, actually, Xn > .J2 (n ?:. 2).
(Use mathematical induction, plus the fact that a 2 + b 2 > 2ab unless
a= b.)
4. Let x = .J2; then x = 2/x. Investigate the usefulness of the recursion
formula xn+l = 2/x.. for calculating .J2.
1.3 Properties of the Real-Number System 5
*5. Let a> -./2. Define a' (Figure 1.1) to be the intersection with the x-axis
of the line tangent to the curve y = x 2 - 2 at the point (a, a 2 - - 2). Find
a'. What does this have to do with Formula (1.6)? (This construction is an
example of "Newton's method," described in most calculus texts.)
Figure 1.1
*6. Use Newton's method to find a recursion formula for calculating iY2.
1.3 Properties of the Real-Number System
In the previous section we tacitly assumed that there is a number denoted by

/2, or, in other words, that there is some (positive) number x such that x 2 =
2.
Indeed such a "number" exists, but, as we now show, x cannot be a rational
number, that is, not a "fraction" p/q with p and q integers. Consequently, the
number system of ordinary mathematics must contain "irrational" numbers.
Theorem 1 There is no rational number whose square equals 2.
Proof Suppose to the contrary that x 2 = 2 for some rational number x = pfq.
Certainly we can assume that p and q have no common factors. The proof will
be completed by "reductio ad absurdum." Namely, we will show that if
(p/q) 2 == 2, then both p and q must necessarily be even integers, that is, p and q
have a common factor of 2. or(c{: ( r''' r ,. 1 -, -::: UH v i I :: o d ci
First note that p 2 = 2q 2 HenJ p 2 is an even integer. Since the square of an
odd integer is always odd (why?j, this shows that p itself is even. Let p = 2k.
Then we have 4k 2 = p 2 = 2q 2 , so that q2 = 2k 2 But, as before, this implies
that q must also be even~ This completes the proof. I
* Starred exercises are somewhat more difficult than average.

The existence of irrational numbers was one of the most confusing facts
facing early mathematicians. Only in the nineteenth century was a successful
theory developed, in which the entire real-number system could be defined in
terms of the system of positive integers.t Since this theory is quite complicated,
it is not convenient to present it in an elementary book. Instead, we will treat
the properties of the real-number system as axioms, referring any interested
reader to more advanced treatises which prove these "axioms" as theorems.t
As far as calculus is concerned, the whole theory may be derived logically from
these properties of the real-number system.
Henceforth, we use the symbol fJl to denote the set of all real numbers.
The notation x E fJl means that xis a member of fJl, that is, xis a real number.
The symbol=> means "implies."
P_roperties of fJl
1. Addition: x,y E f1l=> X+ y E fJl.

2. Multiplication: x, y E fJl => xy E fJl.
3. Commutative Laws: For x,y E fJl,
(a) x + y = y + x, (b) xy = yx.
4. Associative Laws: For x,y E f!4,
(a) x + (y + z) = (x+ y) + z, (b) x(yz) = (xy)z.
5. Distributive Law: x(y + z) = xy + xz.
6. Zero: There is a number 0 E fJl such that
x + 0 = x for every x E fJl.
7. Subtraction: Given x E fJl, there is a unique number
-x E f!4 satisfying x + (-x) = 0.
8. One: There is a number 1 E f!4 such that 1 # 0 and
x l = x for every x E f!4.
9. Division: Given x E f!4, x # 0, there is a unique number
1
x- E f!4 satisfying xx-1 = 1.
t The famous remark "God made the integers; all the rest is the work of man," is at-
tributed to L. Kronecker (1823-1891).
t See, for example, W. Rudin, Principles of Mathematical Analysis, 2d ed;, McGraw-Hill
(1964).
1.3 Properties of the Real-Number System 7
10. Order:
(a) Given x, y E &1, then x = y, x < y, or y < x; no two can hold
simultaneously.
(b) Ifx < y andy< z, then x < z.
(c) Ifx < y, then x + z < y + zfor every z.
(d) Ifx < y, then xz < yzfor every z > 0.
11. Completeness: (See Section 1.8).
We suppose that you are quite familiar with Properties 1 through 10, and
these properties will be used without comment. Notice that Properties 1 through
10 cannot possibly describe &I completely, because the system _1 of all rational
numbers also satisfies Properties 1 through 10. The same holds for certain other, 1(1 ~ u
number systems (see Exercise 4 below). It is therefore interesting that a single \\alfl
additional property, called completeness, is sufficient to describe &I precisely. ~<0...
This property is somewhat subtle, and we will discuss it in detail in Section 1.8. '
If x E &11 then lxl (the absolute value of x) is defined by
X if X;:;::: 0,
lxl = max (x, -:X) = {
-x ifx < 0.
The following simple properties of the absolute value are easily checked:
x :s;; lxl,
(1.8)
lxl = JXI,
lxyl = lxllyl.
To conclude this section we derive a very important inequality, the "tri-

angle inequality." It is an instructive, nontrivial exercise to fill in every logical
detail of this proof, exhibiting every last application of Properties 1 through 10
(Property 11 not being required in the proof).
Theorem 2 (The triangle inequality) For any x, y E &I we have
lx + yl :s;; lxl + lyl. (1.9)

Proof Since both sides of the required inequality are nonnegative, it is sufficient
to prove that
lx + Yl 2 :-:::;: (lxl + lyl) 2

But we have (using (1.8) several times):
lx + yl 2 = (x + y) 2
=x2 +2xy+y2
:-: :;: lxl 2 + 2lxyl
+ lyl 2
= lxl + 2lxllyl + lyl

2 2
= (lxl + lyl) I 2

Let us write the triangle inequality (1.9) in a different form. Since a - b =

(a -c) + (c -b), we see that ~ \;
. I\
la-bl :c;:ja-cl+lc-bj. 0\ /.._ , (1.10)
-.~
G
Definition IJ.a, bE 9e, we define the distance between a and bas
dist (a, b)= Ia- bj.
For example, dist (3, 7) = 13 - 71 = 4, whereas dist ( -3, 7) = l-3 - 71

= 10. Inequality (l.IO) therefore asserts that, for any point c, the distance (rom
a to b is not greater than the sum ophe distances from a to c and from c to b.
This helps to explain why Inequality (1.9) is called the "triangle" inequality.
The triangle inequality also holds for vectors (see Chapter 5), in which case
actual triangles are involved.
Exercises
1. Show that V'2 is not rational.

2. Show that .j3 is not rational.
3. Let a oj:: 0 be a rational number and b an irrational number. Show that
a+ b and ab are both irrational.
4. Let ~ denote the set of all real numbers of the form
a+ bJ2,
VI The Limit of a Sequence (Introduction) 9
where a and b are rational numbers. Show that .fF has all the algebraic
properties 1-10. (Properties 3, 4, 5, 6, 8, and 10 are obvious; concentrate
on Properties I, 2, 7, 9.)
5. Show that the following inequality is valid for all real numbers a, b:
llal - lbll ~ Ia - bl.

(First substitute x = a - b, y = b in the triangle inequality.)
6. When is it true that lx + yl = lxl + lyl?

n n
7. Prove by induction: II X; I ~ L lxJ
k=l k=l
*8. Let .fF be the set of numbers given in Exercise 4. It is obvious that .fF is
larger than the set f2 of rational numbers, since .J2 E .fF but .J2 2.
Show in turn that the set ~is larger than .?F. (Show that .J3 is not in .?F.)
1.4 The Limit of a Sequence (Introduction)
This section gives an intuitive introduction to the concept of the limit of a

sequence. (Other types of limits are discussed in later chapters.) The intuitive
idea is quite easy to grasp, and certain elementary calculations can be carried out
without difficulty. On the other hand, the precise logical definition of limit,
given in Section 1.6, is quite difficult. One cannot hope to proceed very far in
mathematics, however, without mastering the concept of a limit in its exact
formulation.
First then, for the intuitive idea: suppose {x.,} is a given sequence. A
certain real number a wm he called the limit of the sequence {x,.} if "x, becomes
arbitrarily close to a as n becomes large." If this is the case, we write
Iimxn =a,
n-+oo
or, alternatively,
Xn __.. a as n __.. oo;
this last line is read: "xn approaches (or converges to) a as n approaches infinity."
Example 1
Let xn = nf(n + 1). The terms of this sequence are

1 2 3
2' 3' 4' ... '
and it is easy to guess that lim xn = I. This can be made even more transparent
by writing n-oo
Xn = n :
1
= 1 -- C~ 1
) --.. 1 - 0 = 1 as n --.. oo.
Example 2
Let Xn = ( -l)n, that is,
-1, +I, -1, +I, . . ..
It should be clear that this sequence has no limit. (Some students may con-
jecture that the sequence has two limits, +I and -I. But neither a = +I nor
a = -1 satisfies the requirement that "( -l)n becomes arbitrarily close to a as
n becomes large." We will prove in Section 1.6 that a given sequence can have
at most one limit)_
~ Example 3
7}\,
Let Xn = -\Yn. A few terms of this sequence can be calculated approximately:
1.0, 1.41, 1.44, 1.41, 1.38, 1.34, .. . .
At this stage it is far from obvious whether limn-co xn exists, and if so, what its
value is. We will return to this example also in Section 1.6.
The limits of a large class of sequences like that of Example I can be found
by using a simple device:
Example 4
Letxn = (2n 2 - 5)/(n2 + 4n). If we divide both numerator and denominator

by the largest power of n appearing, we get
2-i2
n 2-0
x = -----..-- = 2 (as n--.. oo),
n 1+,1 1+0
n
1.4 The Limit of a Sequence (Introduction) 11
or, in other words, limn-co xn = 2. If this calculation is fairly obvious to you,

then you already have a good intuitive idea of the limit concept.
Example 5
Let Xn = n2 /2n. It is not at first obvious, but we can show that limn-oo Xn = 0.
By the binomial theorem,
2n = (1 + 1)n
= 1 + n + n(n - 1) + n(n - 1)(n - 2) + ... + 1
2 6
n(n - 1)(n ~ 2) or '! 'tl(f\-11 0( 7 \t\

> 6 \ .
}_
Therefore, if n > 2, N {'d 01\~ 1 :; 0
,\1-- \}~'(\ "' n2

<
6n 2 6n ~ -------
/h.
f... ~... ~s l -
2n n(n - 1)(n - 2)
==
n2 - 3n +2,- \ 1 -+ t-
-.,..... -'L-
1-. 'A
which, by the method of Example 4, approaches zero as n ~ oo. Consequently
xn~o as n~ oo.
The question now arises, if an intuitive understanding of limits can be used

to calculate answers to problems, why bother with a more formal logical
approach? There are at least two responses to this question: first, one's intuition
may not be strong enough to handle more difficult problems (such as Xn =
'iYn); second, two people's intuitions may give different answers to the same
problem.
t Example 6
Let xn = C: f. --') ;~ 1
Since we know that nf(n + 1) ~ I, we can guess that xn ~ I as n ~ oo. But

this answer is incorrect, and in point of fact, it is known that limn-co Xn =
1/e'"'"' 0.432. We put this example aside, together with Example 3, to return to it
after a more careful study of limits. (See Section 1.9.)
Exercises
1. Decide which of the following sequences have limits, and find them by
inspection when appropriate.
(a) {n ~ 2} (b) { n(n

4n(n
+ 4)}
+ 1)
(c) {n2 + 6} (d) {sin n
11
},
n+6 2
{~sin ~ },
11
(e)
2. Calculate.
. 100- n (b) lim 2 n n +

(a ) I1m ,
n-+oo 100 + n n-+oo 2n- 5'
na 2n
(c) lim-, (d) lim-,
n-+oo 2n n-oo nn
. 2n
(e) I1m
+ n2 .
n-+oo 3n + n2
3. If Xn = 0.99 ... 9 (n decimal places), what is limn-oo Xn?
4. Concerning Problem 3, students sometimes ask the following question:
"Supposing xn---+ a as n---+ oo, what happens when n = oo ?" How does
one answer this? Is it reasonable to say that a is the "last term of the
sequence"?
5. Let Xn = Jn + 1 - J;,. Show that limn-oo xn = 0. (Hint. Show first that
6. Find
(a) lim (Jn 2 + 2n + 1 - n), (b) lim(Jn 2 + n- n) ..

n-+oo n-+oo
(Neither limit is zero.)

1.5 The Limit of a Sequence (Numerical Examples)
Keep in mind that the formula limn~oo Xn = a is supposed to mean roughly

that xn becomes "very close" to a as n becomes "large." In this section we show
how to answer the question: How large must n be in order to make the distance
from xn to a less than some given "tolerance." Remember that the distance from
Xn to a is equal to
Example I
Let Xn = (n - 1)/(n + 1), so that, obviously, lim 7, _ 00 Xn = l. Given a "toler-

ance" of 0.01, find an integer N such that
Jxn - II < 0.01 for all n ;;:::: N.
Solution By elementary algebra,
Jxn- 1J = n-n + 1- 1 I
I 1
=ln:11
2
n +1
Now 2/(n + 1) < 0.01 = 1/100 is valid if n + 1 > 200. Let N = 200. We have
shown that
Jxn- lJ < 0.01, provided n ;;:::: N.
Example 2
Clearly limn~oo l/(n 3 - 3n + 1) = 0. Given a "tolerance" of I0-3 , determine

an integer N such that
In 3
-
1
3n +1
I< 10- '3
provided n ;;:::: N.
Solution We simplify the denominator as follows:
n3 - 3n + 1> n 3
- 3n
~ 9n- 3n ifn ~ 3 (why?)
= 6n.
Therefore,
In 3
-
1
3n + 1 I< ..!.
6n
if n ~ 3.
/0 ;/t? J
Since 1/6n ~ I0-3 is valid if n ~ 103 /6 : which in turn is valid if n ~ 200, say,
we can takeN= 200. M VI -::;/ / 01
C"l ]1 ')
Example 2 is worthy of careful study. Notice that we do not usually ask for
the smallest possible value of N when dealing with limits. Keeping this in mind
often allows much simplification in calculations. The following lemma is often
useful in this respect.
Lemma Let b = constant. Then

L '/
(i) n + b ~ 2n if n > b,
(ii) n- b ~~ if n ~ 2b. <J ~,, b --::-') h- ~ ~ b )

7
"
.A <r :..__+ C v .:; /V\) " 1

1
r: J 1 h- ~ ~ h Tflj/
The proof is completely trivial. r ' 1
rf' "' ( s l..
Example 3
Take a "tolerance" of 1/100 and find N, as before, for the case
lim
n-oo
(-f!---
n - 7
+ 5n + 1) = 5.
n+5
Solution To simplify the algebra we treat each fraction separately. Observ-

ing that nf(n 2 - 7)- 0, we obtain JfrM~~ 1 -:::::) Vl'- >/ l <-[ ::: 2 . 7 =)
/)\ '- 1 > h'l..-
I 71 ~ n~2
1 "' - / -
if n2 2 14 or if n 2 4, Z-
n2 :_
2 1
=-<- provided n 2 401 = N 1 /
n 200
Next
5n + 1 _ 5 1 = _1!_
1 n+5 n+5
24 1
<- ~- provided n 2 4800 = N 2
n 200
Let us now choose N = max (Nl> N2) = 4800. By the ~e inequality we

conclude that for n 2 N,
_n
l ~+7
+~-5~~~-n
n+5 ~+7
1+1~-5~
n+5
1 1 1
< 200 + 200 = 100 .
Example 4
Observe that limn....oo 2nf(n + 3) = 2. Let e be an arbitrary positive "tolerance."

Determine an integer N (depending on e) such that
I~
n+3
- 21 < e whenever n 2 N .
Solution We proceed as in the previous examples:
1 ~-21--6
n+3 n+3
<~< e provided n > ~.

n e
Hence , if we let N be any integer> 6/e, we are through. (Notice that the smaller
e is, the larger N must be. This is typical of limits of sequences.)
Exercises
I. For each of the following sequences {xn}, find the limit a by inspection.
Then determine some integer N such that
1
lxn - al < - - for every n > N .
1000 -
(a) xn = ( ~~r (Note : IC -1tl = 1.),
(b) Xn = -n+6
2
--,
n -6
(d) Xn = 3na- n + 8 '

l)(n - 2)
n(n -
(e) Xn = (0.9t (first find some k for which xk < 0.1).

(g) Xn = J;;2+l - n,
104
(h) Xn = o1 6
n -10
2. Let e > 0 be an arbitrary (unspecified) tolerance. For each example (a)- (d)
of Exercise 1, determine an integer N, depending one, such that
lxn - al < e for every n ;;:::: N.
3. Use the known fact that sin x < x for x > 0 to find an integer N such that
7r 1
I cos- - 1I< -- for all n 2 N.
n 1000
1.6 The Limit of a Sequence
We come now to the heart of the subject matter of Chapter I. First we will
give the exact definition of the limit of a sequence. Then we will check, by means
of simple examples, that the definition agrees with our intuitive understanding
of limits. In the next section we will derive as theorems some important con-
sequences of this definition.
Students of elementary mathematics frequently wonder why mathe-

maticians are so fussy about giving precise definitions. In elementary courses, the
basic concepts can usually be understood intuitively, and all problems e)1-
countered in the course can be solved on the basis of this intuitive understanding.
This is especially true of a first course in calculus. But in studying more ad-
vanced mathematics, most students eventually come to a stage where they can
no longer really follow the course unless they take the time to understand com-
pletely the underlying ideas of the subject. In fact, many students never seem
to get over this "hump," and consequently, in a state of confusion, abandon
mathematics. I hope that readers of this book will be able to continue their
mathematical education unhindered by any confusion or misconception about the
meaning of limits.
Suppose we know that Jimn~oo Xn = a for a certain given sequence {xn}.
Let c > 0 be a given "tolerance" Then, as in the preceding section, it must be
possible to determine an integer N such that !xn - a! < c for all n 2: N. This
idea is precisely what is taken for the definition of the limit.
Definition 1 Let {xn} be a given sequence of real numbers, and let a be a

given real number. Then
lim Xn =a
n-+oo
means that:
for any given c 2: 0, there is a corresponding integer N (which may depend
on c) such that
!xn - a! < c for every n ;;::::-N . (1.11)
In understanding this definition, it is important to remember that lx~ - al

respresents the distance between Xn and a. The inequality lxn - al <::: c is
;(// ////;'//;{/////;'/;'/;( ..
a-e a
Figure 1.2
equivalent to the double inequality (see Figure 1.2)
a- c< Xn < a c. (1.12)
The above definition can, therefore, also be worded as follows: limu 7 oo xu = a

means that whatever intervalt (a - e, a e) is taken (with e > 0), all the
terms of the sequence {x.. } except possibly the first N - 1 (depending on e)
lie within this interval.
Example 1
lim .! = 0.
n-+ 00 n
To show that this is the case, we consider a given e > 0. We must deter-
mine N (depending on e) so that
-(
I;- 0 I = ~< e for every n 2 N. - t_
Now 1/n < e, provided n > 1/e. Let N be any integer larger than 1/e. Then,
indeed,
-1 ::;: -1 < e if n 2 N.
n N
Example 2
1
lim - - = 0.
n-+oo Jii
To see this, we note that 1/Jfz < e, provided n 2 N > 1/e2
It is worth pausing to examine the form of the proof adopted in Exa:mples
1 and 2. In order to prove, in any specific case, that Iimn=tro x, = a, the follow-
ing steps must be taken:
(i) An arbitrary e > 0 is considered;
(ii) An integer N is determined, by some form of mathematically logical
reasoning, such that Condition (1.11) is satisfied.
The degree of difficulty in any particular example will depend on how compli-
cated the determination of N is. The next example is considerably more difficult
than the previous two.
Example 3
lim \Yn = 1.
n-+ oo
t By the interval (h, k) we mean the set of all numbers lying between hand k exr.bsively,
that is, the set of all x E fit satisfying h < x < k.
.. .,._ - ..
_{ - ~ t
As befoJe-;-len~t~ 0 be given. Let us write Yn = \Y'n- I; then we must
show tha(t-"IYnl < e fo n::::..: N. ~ 5. _~
By th~i theorem (Appendix II), we have, since Yn:;:::.,: 0, r~ 1 1 ZI
7r <
~ \ \ j I
~VI~/ -::.)
n(n - 1) 2 n
vrh // /1(1 :
= 1 + nyn + _ 2 Yn + + Yn ~) ~h::. ~;, .; .J
n(n- 1)
I cun '1

(.y'))f,r. ~()tor ~-(l i{)
> 2
Yn (If n :;: :. : 2). ~ 2 'L
2 'f.-\
> l.f7) ~ 0
Therefore
-- /,{
_
0 ::::;: Yn <
~~
J_2_ . L
n -1
,
i0';\Wt.;
., )
). ~
""~
-;;...) h 1- .,- 1 -"

()A f l(f -J
< __ -=-';
Yl- l
Vl 7 '( 2--
-;;-. !'l
But ../2/(n- 1) ~ ../4jn if n:;:::.,: 2:and this will be less than e if n > 4/e 2
Hence, if N is any int~ger greater than 4/e 2 (and N :;: :. : 2), we have, as desired,
IYnl < e for all n :;: :. : N.
We will now prove that the limit o(a sequence is unique, if it exists at all.
You may object that this is obvious because if Iimn~ao Xn = a and Iimn~ao Xn =
b, then a and bare equal to the same quantity, and consequently a= b. We
challenge you to find the flaw in this argument. It may not be obvious; see

Exercise 5. - ~vv-'":1~-e;.

"~""': (' vL
.. / 1
, , "' :) - ~ 0I -: 1
- CN.; J vlvr c-.:~ J2Jc \ r I..ern C( I
Theorem Jt a given sequence {x.,} converges, then its limit is unique. (In
other words, a sequence cannot have more than one limit.)
Proof Suppose {xn} converges to both a and b. Then, given any e > 0, there
must exist an integer N 1 such that
lxn - al < e for every n :;: :. : N 1
and an integer N 2 such that
lxn - bl < e for every n :;: :. : N 2

20 Chapter One Sequences, Limits, and Real Number:;
Then, by the triangle inequality, we have
Ia - hi = Ia - Xn + Xn - hi
~ Ia - xnl + lxn - bl < 2e, if n 2 max (N1 , N 2 ).
Therefore, given a_EY e > 0, we can show, by choosing a suitable value of n,

that Ia - bl < 2e. But this is possible only if Ia - bl = 0, that is, a= b. I
Definition 2 ,A sequence that does not converge is said to diverge.
Example 4
The sequence {( -1 )n} diverges.

Suppose, on the contrary, we had limn-co ( - I)n = a. Then for every posi-
tive e we would have
1(-l)n- al < e
for all sufficiently large n. Taki!lg a (large) even value of n, we conclude that
II- al <e.
But this inequality must be valid for every e > 0, and therefore a = 1. On the
other hand, the same reasoning using an odd value of n leads to the conclusion
a = -I. This contradicts uniqueness oflimits, and shows that {( -l)n} diverges.
The proofs just given for the theorem on uniqueness and for Example 4,
although very simple, contain some subtle logical points. In both proofs, for
example, we used the logical principle that if a .statement Sn is true for every
value of n in some range (such as n 2 N), then it is true for any particular value
of n in this range.
We also used, in each proof, the mathematical principle that if lxl < e
for every positive e, then x = 0. This intuitively obvious fact can easily be
derived from the axioms for f!J; see Exercise 6.
Exercises
1. Show, by a proof similar to Example I, that
1
lim - - =0.
n-+co 2n - 9
1.7 Elementary Theory of Limits 21
2. Prove that
. 2n 5 +
(a) It m - - - - 2 (b) lim (;l)n = 0,
n-+oo n + 2 - ' n-+oo n + 3
(c) limj(1 -
n-+oo
_n!) = 1,
n2
(d) lim - = 0 (cf. Example 5 of Section 1.4),
n-+oo 2n
2
1 1 2n
(e) lim n / (f) lim -;; = 0,
n-+oo 2n - 5 2' n-+oo 3
(g) lim 2n 3n = 0. +
n-+oo 3n- 2n
3. Each of the following sequences either diverges or converges to zero .

.., Decide which is which and give the complete (''t:-N") proof in each case.
(a) {n}, (b) {( -1~2 + 1}
(c) {~sinn}. (d) {sin n 7T}

3
4. Does a constant sequence (xn = b for all n) converge? Prove your answer.
n 7 / I ---'-"') I -x ., - 6 I -=-- o c.. [: v
5. Read the paragraph preceding the Theorem of this section, and criticize
the argument suggested there for proving uniqueness of limits.
6. Prove that if Jxl < e for every positive e, then x = 0. (Hint: if x =I= 0, let
e = tJxJ and derive a contradiction.) 1)( i < . 1 !?L 1 -:::.) 1 <. ~ :f
~ 1 '
1.7 Elementary Theory of Limits
You will probably agree that the definition of limit given in the previous
section is rather unwieldy. Consider, for example, the intuitively obvious formula
(1.13)
it should not be necessary always to prove such formulas by "t:-N" arguments

such as those given above. What is needed instead is a number of simple general
principles which can be used in dealing with limits. The most important such
principle is the following theorem, which can be summarized by saying that all
algebraic operations are preserved by the limit operation.
Theorem 1 Let {xn} and {yn} be convergent sequences, with
lim X 8 =a and lim Yn =b. (1.14)

n-+oo n-+oo
Then
(i) lim(xn + Yn) =a+ b.
n-+oo
(ii) lim XnYn = ab.
n-+oo
iii) lim~=~ if b :i= 0 and no Yn = 0.

n-+oo Yn b
Before proving this theorem, let us show (in awesome detail) how it is
used in practice on the exampfe (1.13) given above:
6
2+6 1+-;
lim n =lim _ _n_ (by algebra)
2
n-+oo 2n + 1 n-+oo 1
2+-
n2
6
lim (1 + 2)
= n-+oo n (by (iii))
lim ( 2 +
n-+ 00
..!..)
n2
(the limit in the denominator is not zero, as we shall see below). But now
6 6
lim (1 + ) =lim 1 +lim (by (i))
n-+oo n2 n-+oo n-+oo n2
= i + 6(lim
n-+oo
!)n 2
(by (ii) and
Exercise 4, Section 1.6)
=1. (by Example 1, Section 1.6)
Similarly since limn--.oo (2 + Ifn 2) = 2, we get the answer (1.13) at last. Of

course, this whole calculation is trivial and can be done in one's head; such may
not be the case in more complicated examples, however.
Before giving the proof of Theorem I we introduce a simple but useful

lemma. A se uence x is said to be bounded if there is some constant M such
that lxvl < M for all n. For example, {(-I )n} is a boun
is not.
Lemma ,Any convergent sequence is bounded.
Proof Suppose that limn-oo xn = a. Taking e = 1 in the definition of limit,

we see that there exists a fixed integer N such that
lxn - al < I if n ;;::: N.

By the triangle inequality
Define M = max (la1 l, la 2 1, ... , laN- 1 1, I + lal).t Then we have lxnl :::;; M
for every n. I ('L ti 1-x c.\ -:~ N -1 1
We now pass to the proof of part (ii) of Theorem I, and ask you to try to
prove parts (i) (which is a little easier) and (iii) (a little harder) yourself. Before
writing out the formal "e-N" proof, we try to see why (ii) must hold. From the
hypothesis (l.I4) we know that xn -+ a andy n -+ b, or, in other words,
I. lxn - al is small if n is large, and
2. IYn - bl is small if n is large.
We have to prove that XnYn-+ ab; in other words,
3. lxnYn - abl is small if n is large.

To do this we proceed as follows:
= i(Xn- a)yn + a(yn- b)l

:::;; lxn - al (.{;,nl + lallyn- bl. (1.15)
-/ Ke.,
Now we see that the terms on the last line are indeed small for large n. The formal
proof follows.
t Note that a finite set of numbers certainly has a maximum.

24 Chapter One S"'qllences, Limits, and Real Numbers
Proof of Theorem 1 (ii) First, by the Lemma {y11 } is bounded; in other words,
there is a positive number M such that
IYnl s M for all n.
Now let e > 0 be given. By the definition of limit, there is an integer N 1 such
that t
e
lxn- al < - for all n 2 N 1 (1.16)
2M
Also, if a i= 0, there is an integer N 2 such that
e
IYn- bl < - for all n 2 N 2 (1.17)
2lal
Let N =max (N1 , N 2) if a i= 0, and let N = N 1 if a= 0. Then by (1.15) we
have, for all n 2 N,
e" e
<-M+Iai-
2M 21al
=e
(the term lal e/2 lal is omitted in case a = 0). This completes the proof of
Theorem l (ii). I
Constructing "e-N" proofs like this is rather difficult at first, but such
proofs occur frequently in mathematical analysis.
Another important fact is that order ( s) is preserved by the limit operation.
Theorem 2 Let {x11 } and {y11} be convergent sequenfes, and suppose X 11 < Yn
(or every n. Then
lim X 11 slim Yn-
n-toco n-+oo
t It is important to understand the logic at this point. The symbol "e," which has already
been specified in this proof, does not now represent the same quantity as the "e" used in
(1.11) of the definition. Rather the "e" of Equation (1.11) is now replaced by the quantity
ef2M (which is a positive number). The same remark applies also to (1.17). Note also, however,
that the integer N referred to by (1.11) depends one in general, and hence is not necessarily the
same in (1.16) and (1.17). We have taken care of this possibility by using the notation N, and
N 2 respectively.
-o- - - - r-~- -tf-

v ~ 0. fN {!-._ f
Proof Let e be a given positive number. Then there is an integer N such that
lxn- al < e and IYn- bl < e for n ~ N,
where a and bare the limits of {xn} and {yn}, respectively. This implies that
0 /)\-~ <( :X t~ t 1 rt~ < b" (
( a- Xn < e and - Yn- b < e for n ~ N
(explain!). Consequently, for n ~ N,
a - b = (a - Xn) + (yn -b) + (Xn - Yn)

< 2e,
since Xn- Yn ::;;: 0 by hypothesis. We have shown, therefore, that a- b < 2e

for any positive e. This means that a - b ::;;: 0, or
a =lim Xn :S::: b =lim Yn- I

n-+w n-+oo
It should be noted that strict inequality (xn & n) is not necessarily pre-
served by the limit operation: see Exercise 10.
Next we calculate two especially important limits.
Theorem l
,-:- l / :"' ('

I ' r- y
( ,: ~) L. !
(i) I(p > 0, then lA

r I c(.
(
lim_!_= 0. / \
)
) <;
n-+oo nf) /
(ii) If Ia! < 1, then
----+--- -f) __
lima"= 0. (1.18)
n-+co
-\ 0\ 0 l
Proof The proof of (i) is easy, and is left to the reader. To prove (ii) we note
that if Ia! < I', t~en 1~1 = -I--

1>/t' '1 :
1
~;:,,,).)1' +q
for some q 30. Now, by the binomial theorem,
(I + q)" ~ I + nq > nq,
and therefore
!an - Ol = !ani = I < _!_ < e if n > _!_.

(1 + q)" nq eq
This proves (ii). I

Finally, we calculate a useful limit of a different sort. The sum
1- an+l
1 + a + a + + an =
2
1-a
(a =F 1) (1.19)
is sometimes called a geometric progression. From Theorem 3 (ii) we obtain
1
lim(1 +a+ a 2 + +an)= - - (iflal < 1). (1.20)
n-+oo 1- a
Formula (1.20) is usually expressed in the notation of infinite series (see Section
1.12) as:
00 1
_Lan = -- iflal < 1.
n=O 1-a
Exercises
1. This exercise is designed to provide you with some practice in the use of the
logical phrases "for every (all, each)" and "for some" (or "there exists").
A more detailed treatment of these logical quantifiers, as they are called,
is to be found in Appendix I.
The following 5 statements refer to a sequence {xn} and a real QUmber a.

The symbol N denotes a positive integer, and e a positive number. For
each statement (a)-(e), determine for which of the sequences (1)-(7) it is
true.
Statements
(a) For every 8 > 0 the inequality lxn - al < 8 holds for all but finitely
many n.
(b) For every 8 > 0 the inequality !xn - al < 8 holds for infinitely many n.
(c) For some N, the inequality lxn - al < e holds for all n ~Nand for all
e > 0.
(d) There exists an 8 > 0 such that !xn - al ~ 8 for all sufficiently large
values of n.
(e) For every N there exists an 8 > 0 such that lxn - al < 1/N whenever
n ~ 1/8.
Sequences
(1) Xn = n -18- n!; a= -8.
(2) Xn = 1 + (-l)n ; a = 1.
n
(3) Xn = (-l)n + ! ; a = 1.
n
(4) Xn = 3; a = 3.
(5) Xn = (1 + ~r; a = 2.
(6)xn=n-18-n!; a=8.
(7) Xn = 2nfn 2 ; a = 1.
2. Prove that if limn-+oo Xn =a, then limn-+oo lxnl = Ia!. (Hint: llxnl - !all
~ lxn -a! by the triangle inequality.)
3. Suppose limn-+oo xn =a and xn > 0 for all n, and a> 0. Prove that
limn---oo ~ = J a.
4. Prove that limn-+oo cxn = c limn-+oo Xn, (Prove this directly, but note that
it is a special case of Theorem 1(ii).)
5. Prove Theorem l(i). (Hint: lxn + Yn- (a+ b)!~ lxn- a! + IYn - b!.)
6.Prove the following Lemma: If limn---ooYn = b =I= 0, then IYnl :?:: ! lbl
for large n (i.e. for all n :?:: N 0 for some integer N 0 ).
7. Suppose that limn-+ooYn = b =I= 0, and no Yn = 0. Use the lemma of
Exercise 6 to show that limn-+oo 1/yn = 1/b.
. 11Yn - b11 =
(Hmt: IYn - bl )
IYnllbl .
8. Prove Theorem 1(iii). (Hint: Combine Exercise 7 with Theorem 1(ii).)

9. Prove by the method of Theorem 3(ii) that limn-+oo nan = 0 if Ia I < 1.
fiD.'l Let {xn} and {yn} be convergent sequences, and suppose xn < Yn for every
~ n. Does it follow that limn-+oo Xn < limn-+oo Yn? Why?
1.8 The Completeness Property
Given a sequence {xn}, it is occasidnally possible to show that it converges

and to calculate the limit exactly by simple methods. But in many practical
problems the reverse happens-namely, a sequential method is devised in order
to calculate some otherwise unknown number. The question then arises whether
the sequence obtained does converge, and if so, whether it converges to the
desired number. (Also often of practical interest is the question : How fast does
the sequence approach the desired number?)
Example 1
Let {xn} be the sequence of Section 1.2:
(1.21)
We can show now that if this sequence converges, then it must converge to .J2.
To see this, suppose
lim Xn =a.
n->oo
Then also limn_.oo Xn+l = a, so that by (1.21)
. . x;.
+2 2 +2
a =hmxn+I =hm - - = - - -
a
n->oo n->oo 2xn 2a
Solving for a, we find a = ../2. Since xn ~ ../2 for all n (see (1.7)), it must be
the case that a= ../2, as asserted.
To complete the discussion of the sequence (1.21) we only have to see why
it must converge. First we recall from Section 1.2 the following facts:
(i) Xn ~ .J2 for all n > 1,

(ii) Xn+l ~ Xn for all n > 1.
Any sequence {xn} satisfying condition (ii) is said to be nonjncreasing. The
sequence {xn} of (1.21) can be pictured as in Figure 1.3.
11111111111 I I
X,
Figure .1.3
A basic property of the real-number system &I, which completes our list
given in Section 1.3, is:
11. Completeness Property of &/. Every non increasing sequence of real numbers
that is bounded below converges to some real number. Likewise, {!Very non-
decreasing sequence of real numbers that is bounded above conuerges to mme
real number Ill n
'VV
i' "'""' ,, ''
J,.l fVV
.
The completeness property is extremely important for the logical founda-

tions of calculus (or "analysis"). Notice that the completeness property im-
mediately applies to our sequence {xn} of (1.21), which must therefore converge
to some real number. By what we have already proved, the limit must be J'i:
lim Xn = J'i if Xn is given by (1.21).

n-->oo
Example 2
Consider the sequence

1
Zn = 1+-
2!
+ -3!1 + + -n!1 .
Obviously zn increases with n. And since
1 1
k! 2. 3 . . . . . k /
{ r' "'' [ t -:. r
1 1 - z_- /..} ; 2-
.
:::;: 2 2 2 =k1
_2- r " r'
,- f- ;C+ '
. {J 1,- O~) 1- { ;_) _ 1- (; .S - -
we have \ -t- 1.__;-< --r-I
__
4 ,._.
_
...-- -;::-:::--
1
., !-;f__
z <
1 - 1 1 r
1 +- +- + ... +- = 2 - - < 2
..
n - 2 22 2n-l 2n- l ' 2 .., 4
1-t-'1'+ 2--.."'-+t"~ :r~:l-'"'
-2.;::: 2 -f::. -;,-I
that is, th,e sequence {zn} is bounded above by 2. Therefore limn_,.oo zn = a
2
X
exists, as a consequence of the completeness property. In the next section we will
show that a= e - I ~ 1.718.
Another important example of an increasing, bounded sequence is
this sequence is analyzed in detail in the next section.

The following theorem has already been used in several examples. Some
will think that the result of the theorem is trivial; those who do may enjoy
trying to find a proof which does not depend on the completeness property.
l... !' '3
I \
\---\~) H::; I -~ '
I\, -t-:::)
'
_Theorem (Archimedean order property of Bl) Let X be a real monher.

Then there is an integer n greater than x.
Proof Suppose, to the contrary, that every integer n is :::;:; x. Then the sequence
{xn} given by xn = n is increasing and bounded. Hence limn_.oo Xn = a for some
real number a. Then also limn_.oo Xn+I =a. But
so that
Iimxn+I = 1 +a.
n-+oo
This implies that a = I + a and, consequently, that 0 = I. Since 0 = I is

absurd, the proof is complete. I
It follows from this theorem (and conversely; see Exercise 8) that:
If e is a given positive number, then there exists an integer N such

that IIN < e. ;&r?.t;vu, X -=) ~ " ? : -==.) [. > J
1
(A)
v Jp 7:: f' 1'1
If this property seems trivial to you, try to prove it using only Properties 1
through 10 of Bl.
Next we present a brief discussion of decimals. By definition a decimal
expansion is an expression of the form
where m is a nonnegative integer and each ai is one of the digits from 0 to 9.

Define a sequence {zn} by writing
Obviously {zn} is increasing; but also
9 9 9
zn ::;;m+-+-+
10 102
.. +-<m+1
10n '
so that {zn} is bounded. By the completeness property, z =limn zn exists.

Naturally we write >ee; rio ' l ? , t
z = m. a1a2a3 an . .. ,
and we say that this is the decimal expansion ofz,

Conversely it can be shown that eyecy positive real number bas a decimal
expansion. For example, given z, we define m to be the greatest integer :s;;;z.
Then a 1 is defined to be the greatest integer :s;;; 10 (z - m), and so on. Except for
the possibility of repeating 9's [0.3I999 = 0.32000 ], the decimal ex-
pansion is unique.
It is an interesting observation that the decimal expansion of any rational
number must eyentualJy repeat.t whereas the decimal expansion of any irra-
tional number is nonrepeating It may amuse you to try proving these assertions.
It would be difficult to overstress the importance of the completeness
property in the study of analysis. This property will be used again and again in
this book; in Chapter 3 we study the property itself with some care. I hope that
you feel that.the completeness property is intuitively reasonable. Actually if one
begins with a clear definition of real number, it is possible to prove the complete-
ness property. t However, since we have chosen to take fJ1l and its properties
as a starting point, we accept the completeness property as an axiom.
Exercises
I. If {x..} is defined by x 1 = I and
x~+ b
Xn+l = - - (n :2: 1),
2x ..
where b > 0 is given, show in detail that limn-+> x .. = .J"b.

2. Suppose the sequence {a..} satisfies
an+l = .JI +a.. (n :2: I).
(a) Prove by induction that {a..} is nondecreasing and that a.. < 2 for ali n.
(b) Show that limn-.> a .. exists, and find it.
3. Define a sequence {w..} by w1 = I and
2
W n+l = 2 if n :2: 1.
w..
t A decimal expansion is called repeating if the digits, from some point on, occur in a
repeating group. For example, 2.7142142142 is a repeating decimal, which therefore is the
decimal expansion of some rational number.
t cf. W. Rudin, Principles of Mathematical Analysis, 2d ed., McGraw-Hill (1964).
Show first that iflimn~"' wn = w exists, then w3 = 2. But then show that the
sequence {wn} does not approach any limit. (This shows that one shouldn't
jump to conclusions!)
4. (a) Prove by induction that
1
1+2
2
+ 231 + + 2n1 L
::;: 2 - - (n = 1, 2, 3, ...).
n
(b) What can you say about
5. Show by induction that
1+ -12 + -12 + + -12 ::;: -7 - -1 if n ;;:::: 2.

2 3 n 4 n
Hence, answer Exercise 4(b) more precisely.
*6. Expanding on the ideas of Exercises 4 and 5, show how to calculate the limit
4(b) to within a given error.
1 3 5 (2n - 1)
7. Let xn = . Show that
2 4 6 2n
lim xn = x exists and 0 ::;: x ::;: l

n-+oo
8. Show that the Archimedean order property implies Condition (A) and
conversely,
9. Detect the use of (A) in proving that limH"' 1/n = 0 (see Example 1,
Section 1.6).
10. Show that if a = pfq is a rational number, then there exists an integer
N >a; show this without using the completeness or Archimedean order
properties of &1.
*11. Consider the sequence {xn} defined by
1- x.,
Xn+l = - - -
2
Calculate the first few terms; then discuss convergence of the sequence.
(Hint: Consider the "alternate" subsequences {x2,._1 } and {x2,.}.)
1.9 The Base of Natural Logarithms
Consider the sequence

(1.22)
We are going to prove that {x,.} is a bounded, increasing sequence, which con-
sequently must converge, according to the completeness property. Hence we
can make the following statement.
Definition lim
n_.co
(1 + ~)"
n
=e. The number e is called the base of natural
logarithms.
Calculus students may be familiar withe because of the important formulas
d ~)( 1
-log x = - '
dX X
which are discussed in every calculus text.

/7) We show first that {x,.} is bounded. By the binomial theorem, we have
ciJ l /J~ z.l.
2
1 ,;(,;=~(1)
x,. ( n1)"
= 1 + - = 1 + -n . - + - i+- ... ?.~+ (1)"
-'
-
1! n 2! '!!/ n
1 1 1
<1+-+-++-
1! 2! n! ------ _____. (.
1 1 1)
<1+ (1+-+-++-
- 2 22 2"-1
< 1 + 2 = 3 /(!) (1.23)
(here we used the simple inequality 1/k! ~ 1/2Tc-l). Thus x,. < 3 for every n;
that is, {x,.} is a bounded sequence. N&)we show that {x,.} is an increasing
sequence. By the binomial theorem once more,
Xn = (1 + ;r
= 1 + 1 + n(n - 1) . ..!. + n(n - l)(n - 2) . ..!. + ... + n! . .!._
2 n2 3! n3 n! nn
+ n!1 ( 1- ~1) ( 1- -n-

n- 1) . (1.24)
In the analogous formula for Xn+l every term on the right-hand side increases \ -~1 r 1--t_
in value and an additional term appears on the right. Therefore, xn+l > Xn-
We have now shown that the sequen'Z {xn} is both bounded and increasing;
consequently e = limn__..oo xn exists. V"'
@
(!;)we now derive another formula fore, which will be useful for calculating
e numerically. Consider the sequence {yn} defined by
1 1 1
y =1+-+-+ .. +-.
n 1! 2! n!
Obviously {yn} is an increasing sequence. It follows from (1.23) that Yn <3

for all n. Therefore, by the completeness property,
f=limyn ~
n->oo
exists. We now want to prove that
First, from (1.23), we have

~--,
ref) '--e e. ___)
-
-t/ a ; L'
1)n <1+-+-+1 1 1
X=
n ( 1+-
n 1! 2!
.. +-=y.
n! n
Therefore, ,r, r -D-1) :/' ,
. e =limxn ~ limyn =f. /fr~ (1.25)
n-+oo I n-+co v;-4t.A_)
r ~ . ____.
( !]) Conversely, let m be a given integer. Suppose n ; ;: : m; taking the first m

terms in (1.24) we get
Xn > 2 + ..!. (1 -
2!
!)n + ..!.3! (1 - !)n (1 - ~)
n
+ "
+ m!1 ( 1- ~1) ... ( 1- -m-

n -1) .
Keeping m fixed, we let n ---+ oo on both sides of this inequality, obtaining
e!2::J
L', 2 1 1
+-+-+ + -1= y .
U 2! 3! m! m
Therefore,7m :S:: e for every m, and hence
(1.26)
We conclude from (1.25) and (1.26) thatf = e, or, in other words,
(1.27)
(In Section 1.12 we will write formulas like (1.27) in terms of "infinite series,"
iti this case e =I;;> 1/n!)
{r?J Formula (1.27) leads to an efficient scheme for numerical calculation of e
~,_"'ito any desired degree of accuracy. /(X)
Let us show first that for n :;:::: 1,
(::7
' 1
Yn < e :S:: Yn + - - (1.28)
n n!
If n is a given integer, let m'> n. Then
1 1
Ym = 'Y n \+
1
(n + 1) .1 + ' + -m.I
/
= Yn + 1 [1 + _1_ + 1 + ... + ------=1-----=-]

(n + 1)! n +2 (n + 2)(n + 3) (n + 2) m
< Yn
1 [ 1 1
+ (n + 1)! 1 + n + 1 + (n + 1) + ... + (n + l)m -f.~:-1.:
2
1 J L__
,,r f';lr
, of\~p-
"') f
lv.r
1 1 1 _.,_,;:
< Yn + (n + 1) .I _ _1_ = Y n + -.-1 '

n n.
.ffrl'.l J
1
n+1
wbere we have again used Formula (1.19) for the sum of a geometric progres-
sion. Since Ym < Yn +
1/(n n!) for every m, we see that
. 1
e= ltmym :S::Yn+--,
(""... "' n n!
which proves (1.28).
~ \ \
( J Jv On the basis of formula (1.28) we can devise an "automatic" scheme for
computing e to any prescribed accuracy. Define two sequences {y,.} and {",.}
recursively as follows:
)
/
" = b,_l .
" n + 1
It is easy to check that b,. = I/(n + 1)!, and therefore
y,. = 1
1
+ 1 +- + ... +-'1\
2! n! <
') Y" \
as before. We also define

"'
Cr;.rd.>.
then by (1.28)
y,. < e <y,. + e,._l. /( 'J
Table I shows a sample calculation of e, with an error :::;:; I0- 5 , using the above
_formulas to calculate y,., b,., and e,. in order. The result at stage 8 is: y 8 < e <
y 8 + e7 ; that is,
2.718 278 7 < e< 2.718 282 7.
Table I Numerical Calculation of e
n y,. 5,. e,. is less than:
0 1.0000000 1.0000000 1.0

1 2.0000000 0.5000000 0.3
2 2.500 000 0 0.166 666 7 0.06
3 2.666 666 7 0.0416667 0.02
4 2.708 333 4 0.008 333 3 0.003
5 2.7166667 0.001 388 8 0.0003
6 2.718 055 5 0.0001984 0.00003
7 2.718 253 9 0.0000248 0.000004
8 2.718 278 7
!.10 Growth Properties of Certain Sequences 37
Thus e = 2.71828 to within an error <10- 6 .t Table I was calculated by hand in

less than 10 minutes. Similar methods are known for the computation of values
of the functions&', sin x, and cos x . (Such methods can also be used in automatic
digital computers, but they are not really used in practice, because much more
efficient methods have been invented! By the way, a computer program consists
essentially of just two items: (a) a recurrence scheme for calculating values of a
sequence (or several sequences), and (b) instructions on where to stop.)
Exercises
1. Calculate e to within 10- 6 (Use Table I. What about using Table I for an
error ~ 10-7 ?)
2. (a) Show by induction that
2" 1
- <- ifn ~4.
n! 2"- 4
(b) Prove that lim

n-+oo
(1 + ~)"exists.
n
.
3. Prove that
lim
n-+oo
(1 + ~)"=lim
n
(1 +
n-+oo
2+
22
2!
+ +
2
n!
").
4. Prove that lim,..... oo ( 1 + ~r = e2 (Hint : note that (1 + 2/n)" =

[(1 +1/(n/2))"'2 ] 2. )
1.10 Growth Properties of Certain Sequences
The following notation is quite convenient.
Definition We say that a given sequence {x,.} diverges to oo, or lim,.....00

x,. = +oo.if.
for any given positive number M, there is an integer N (depending on

M) such that x,. ~ M for every n ~ N.
t This does not take into account the possibility of accumulated roundotr errors, which
however cannot exceed 8 x t0- 7 Therefore, we are certain that the computed value of e is
correct to within 10-.
The sequences {n2 } and {211 } are obvious examples of sequences diverging
to + oo.
Theorem 1 Let {x.,} be a sequence of positive numbers. Then
lim x .. =
n-+oo
+ oo if and only if lim _.!_ = 0.
Proof Suppose X 11-+ + oo. Then for any e > 0 we can determine an integer
N such that x .. > I/e when n ~ N. Therefore, Ifx .. < e for n ~ N, which im-
plies (since Ifx.. > 0) that I/x.. -+ 0. The converse is similar. I
It is a well-known scientific fact (which has important consequences in

chemistry, biology, economics, and other subjects) that any exponential
function a'% (a> I) grows faster as x-+ + oo than any power of x. In terms of
sequences, this can be expressed as follows.
Theorem 2 Let a > 1 and p > 0 be given real numbers. Then
an
lim-= +oo. (1.29)
n-+oo n"
Proof Since a > I, we have a = I + q for some q > 0. Let k be any integer
such that k > p. Consider values of n ~ 2k. Then we will surely have
n
n>n-1>>n-k+l>n-k~-.
2
Now we apply the binomial theorem:
n
a =
(l + q)n > n(n - 1) (n - k
k!
+ 1) q,.
Therefore,
1.10 Growth Properties of Certain Sequences 39
where the expression "const" represents some real number (here, qkf2k k!)
which does not depend on the variable n. Since k - p > 0, it follows that
anfnP-+ + oo as n-+ oo. I
Another useful limit is expressed in the following theorem.
Theorem 3 Let a be a given real number. Then
an
lim-= 0. (1.30)
n-+co n!
J
Proof Choose any integer M 2 JaJ. Now if n 22M, we can write
~ lj v-IJ"'
''J.~- L (~
I I : ; Mnn!
an
f.
tl\ J
n!
--,,
MMfl-i)
L
// MMM~
?.1L. I
--------~-- .
MMM
------- ~
=...___,_1 2 ;_ M
1)n-2M
7
....___ ___________ --
(M + 1)(M + 2) 02M)) (2M+ 1)(2M + 2) n J
---~----- '
<MM1M (2
Since M is constant, it follows that Jan/n!J-. 0 as n-+ oo. I
Exercises
1. The following sequences {xn} approach + oo as n-. oo. Determine for

each an integer N such that
Xn > 1,000 for all n 2 N.
(c) xn = GT
nn
2. Show that lim - = + oo.
n-+co n!
3. Assume a> 0. Determine limn..... co {(a+ I)n -an}; give reasons for your
conclusion. (Hint: Consider the cases a< 1 and a 2 1 separately.)
4. Let {x,.} be a sequence satisfying the condition
where A < 1 is a constant. Show that lim,._"" x,. = 0.
5. Suppose that lim,._"" x,. = + oo and lim,._ coy,. = a > 0. Prov_e that
lim x,.y,. = +oo.

n-+oo
6. Given that lim,._ 00 x,. = + oo, prove that
. m-
It x,.- - =1.
1 +X,.
n-+oo
Also prove the converse under the additional assumption that x,. > 0 for
all n.
7. Let k be a fixed integer and let a 1, a 2, ... , ak be constants. Prove that
lim (nk
n-+oo
+ a1nk-l + a 2nk-2 + + ak) = + oo.
Do not assume ai ~ 0. (Hint: Use Exercise 5.)
8. If b > 1, show that
1.11 Some Further Properties of the Real Number System
In this chapter we have stressed a fundamental distinction between the

system 2- of all rational numbers, and the system Pli of all real numbers, namely
the property of completeness. We have indicated the importance of this property
in the study of limits. In this section we wish to discuss briefly some questions
related to the fact that the number systems 2- and Pli (as well as the system .AI'
of all positive integers) contain infinitely many elements. Questions about the
infinite have plagued mathematicians, philosophers, and students since classical
times. Since the pioneering work of Georg Cantor, however, mathematicians
have been able to develop virtually complete mastery over the concept of the
infinite.
Cantor's brilliant contribution to mathematics begins with the following
definition, which tells us under what conditions two sets (finite or infinite) have
the "same number" of elements. Since "number" already had a fairly definite
meaning in mathematics, Cantor introduced the term "equivalent" sets.
Definition I Two sets A and B are said to be equivalent if there exists a

one-to-one correspondence between the elements of A and the elements of B.
Obviously two finite sets A and Bare equivalent if and only if they have the
same number of elements, in the usual sense. Cantor's definition is most in-
teresting, however, when applied to infinite <;ets.
For example, if .;V = {I, 2, 3, ... } is the set of positive integers, let
% 2 = {2, 4, 6, ... } be the set of all even positive integers. These sets are
equivalent, since the corespondence n~ 2n is a one-to-one corn;spondence
between the elements of JV and JV 2
Note that % 2 is a proper subset of .IV. Thus an infinite set may be equivalent
to a proper subset of itself; in fact, this property is characteristic of infinite sets
(see Exercise 4).
Any set which is equivalent to the set JV of positive integers is said to be
countable, or countably infinite (or, sometimes, enumerable). Thus the set % 2
of even integers is countable; it is easy to show (Exercise I) that the set f!l' of all
integers, positive and negative, is also countable. The question immediately
arises whether all infinite sets are countable, and in particular, what about the
sets !2 and &I?
Theorem I (Cantor) The set !2 of all rational numbers is countable, but the
set &I of all real numbers i$ uncountable.
Proof A given set S is countable if and only if there is a one-to-one function

from JV to S, such that each member of S corresponds to a unique positive
integer n. But a function from .;V to Sis nothing but a sequence. Hence we can
say that Sis countable if and only if there exists a sequence {xn} such that every
member of S occurs exactly once as a term of the sequence.
(a) !2 is countable: Consider the following sequence {xn},
--, t, -t, i, - !, f, -f, l, -l, t, -t, ... '

~ L-y--J '------,,-----' '------,,-----'
(I) (2) (3) (4)
42 Cluzpter One Sequences, Limits, and Real Numbers
where the kth group contains all positive and negative rational numbers in
lowest terms, whose numerator and denominator add up to k. Clearly every
rational number pfq occurs exactly once in this sequence.
(b) fJI is uncountable: Suppose to the contrary that fJI is countable, and
let {xn} be a sequence containing all real numbers. Let us write out the decimal t
expansions of each X : ';, v I nUn vw I
n I Nl} JA) r) '. ' /1; "'' otJ.-P--
Xt = nt. auat2ats. . 1"" 1 , {/'
9' ~
(1.31)
and so forth . Here the numbers a;; are digits (0 to 9). According to our assump-
tion, every real number appears on this list somewhere.
Now for each i = I, 2, 3, ... , let bi be any digit except aii, 0, or 9. Con- ..;;-{ .
sider the real number r.f!' ll!v'(J)
x=O.btb2ba -> f;-;, C). (;~4
Since the decimal expansion of xis different from every decimal in the se~~ence ~ I (1
(1.31) (and since x did not arise tl)rough the ambiguity of repeating 9's), we
reach the contradiction that xis not on the list (1.31). This completes the proof.
This method of proving (b) is sometimes called "Cantor's diagonalization
process." I
The proof of the following theorem, which is a generalization of Theorem

1, is left for the exercises.
Theorem 2 Let a < b be two arbitrary real numbers. Then the set of all
rational numbers lying between a and b is (infinite and) countable, whereas the set
of all irrational numbers lying between a and b is (infinite and) uncountable.
Definition 2 Any set S with the property that every open interval (a, b),
a < b, contains at least one point of S, is said to be a dense subset of fJI (see
Figure 1.4).
/ An element of S
a b
Figure 1.4
We see, for example, by Theorem2, that !2 is a countable, dense subset of f!,l .

Theorem 3 If S is a dense subset of &1, then every real number r is the limit
of some sequence {x,.} of members of S.
Proof For each positive integer n, consider the open interval (r- 1/n, r + 1/n).
By hypothesis, there is a member x,. of Sin this interval. Thus lx,. - rl < 1/n,
and therefore, Iim,...... co x,. = r. I
Exercises
I. Show that the set fl' = {0, 1, 2, ...} of all integers is countable.
2. Let W denote the set of all ordered pairs (n, m) of positive integers. Show
that W is countable.
3. LetS and T be countable sets. Show that the unionS u T (consisting of all
members of S together with all members of T) is also countable.
*4. Let A be an infinite set, and let Bbe obtained by deleting one element of A.
Show that A and B are equivalent.
5. Show that any interval (a, b), a < b, is equivalent to the interval ( ....:.1, +1).
Hence, all nonempty, open intervals are equivalent.
6. By means of the function f(x) = xf..Jx2 + 1, show that the interval

(-1, +I) is equivalent to &I (see Figure 1.5).
-I
Figure 1.5_
7. Prove Theorem 2.
8. Let S be a set derived from f2 by removing finitely many elements of fl .

Show that S is a dense subset of &1. Could infinitely many elements be re-
moved in some way, still leaving a dense subsetrof &I?
9. Which of the following subsets are dense in 81?

(a) a finite set~;
(b) the set fl' of all integers;
(c) the set of all rational numbers with even denominators (in lowest
terms);
(d) the set of "dyadic" rationals ;m'
n E fl', mE .IV.
1.12 Infinite Seriest
In this section we study a special type of sequence called an infinite series.

Since most ordinary calculus texts also discuss infinite series, we will only give
a few elementary examples together with some of the basic theory. Infinite
series provide inter~sting examples of convergent sequences whose limits often
cannot be found in simple form. They have a wide range of applications both
to theoretical and practical problems in mathematics.
..,
Let {an} be a given sequence of real numbers. We can define a new sequence
{sn} by forming the sum
Sn = zn
k~ 1
ak = a1 + a + + an.
2 (1.32)
If it happens that the new sequence {sn} converges to the limit S (limn--+oo sn =
S), we write
00
zan=
n~1
s, (1.33)
and we call S the sum of the infinite series z~- 1 an. The sum sn of (1.32) ~
called the nth partial sum of the series~:::' , an. ,
In other words, by definition
oo n
zan =lim zak, (1.34)
n=l n-+ oo k=l
provided this limit exists. If the limit does exist, we say that the infinite series
L~~t an converges (to S). If limn_,oo Sn does not exist, we say that !~~ 1 an
diverges. In the special case that limn--+oo sn = +
oo we say that !~~ 1 an ar-
verges to + oo.
t This section may be postponed until beginning Section 4.6.

Example I
oo I
2- =
n=12n
I. To verify this we calculate the partial sums:
hH
Sn
n 1
= k~l 2k =
1
l
1
+ 22 + + 2n
1 t (~) -=-
1
\ - :2--
1
=1--
2n'
by the formula for summing a geometric progression. It follows that limn--+oo sn =

I , which means the same as 2~= 1 I f2n = I.
Example I will be generalized in Theorem 5 below.
Example 2
2~=1 ( - I)n diverges, because
s1 = -I,
s2 =-I+ 1 = 0,
s 3 = -1 + 1- 1 = -1,
and so forth . Thus the sequence {sn} is
-1, 0, -I, 0, -1, 0, .. . '

which surely diverges.
Example 3 \ Y\
't n
~ \
oo n 1(\11 \
2 - - diverges to + oo. -; 1 If\>;
n=l n + 1 ~Y\
To check this we notice first that _n_, ::;::: l (by algebra) for all n; there-
n +I
fore,
1 2 n 1
sn =-+-++-->n-
2 3 n+1 2'
so Sn - + oo as n - oo.
It seems evident that a series Lf au can't converge unless an_. 0 as n _. oo.

We can easily prove that this is the case.
Theorem 1 If !f' an converges, then limn-c.o an = 0. The converse is false.
Proof Suppose first that !f an converges. Then limn-c.o sn = s exists. Note that
It follows that
lim an+l = lim Sn+l - lim Sn = S - S = 0.
n-+c.o n-+co
Therefore also limn-c.o an = 0.

To prove that the converse is false, we consider the following case, known
as the harmonic series, in which an_. 0 but !fan diverges to oo. I +
Example 4
The harmonic series :f .!n diverges to + oo.

n=l
(1.35)
We can establish this by using some simp!e calculus. Consider the curve
1
Y =f(x) = - (x > 0).
X
Area Y<l
X
2 3 4 5
Figure 1.6
Smce f(x) = Ifx is decreasing, it is fairly obvious from Figure 1.6 that we have
1 1 1 rn+l 1
sn = 1 + - + - + + - > J,1 - dx = log (n + 1).
2 3 n x
Hence, sn- + oo as n- oo, which means!~ diverges to + oo. I

sl~re lo"')(/1-+ 1
-) -'--:> n
~->e:V
The simple method above is a special case of the "integral test" usually
discussed in calcul~s texts. The same method .can also be used to prove that
certain series converge.
Example 5
00 1
!
n~1
2 converges.
n
To verify this, consider the graph of y =f(x) .- l/x2 in Figure 1.7.
2 3 4 5
Figure 1.7
w~ see that
Sn =! n
-1 J,n dx
< 1 + 1 -2 = 2 -
1
- < 2.t (1.36)
k~1 k 2
x n
t We recall that the inequality (1.36) was to be proved by induction in Exercise 4 of

Section 1.8.
Hence, {s,.} is a bounded, increasing sequence, which must therefore converge,

by the completeness property. t
The method used in Examples 4 and 5 can also be used to derive the follow-
ing general result, but we will not pause to give the proof.
Theorem 2 (The hyperharmonic series)
00 1
! -; <
n=l n
oo if and only if p > 1. (1.37)
We next present two simple but important theorems regarding the con-
vergence of series. When combined, for example, with Theorem 2, these
theorems permit one to test many given series for convergence-see Example 6
below.
_'fheorem 3 Le~ be a series o( nonegative terms (that is, a .. ;;::: 0 for

all n). Then the series converges i(and only if the sequence {s,.} ofpartial sums is
bounded.
Proof By definition !r' a.. converges if and only if the sequence {s..} con-
verges. But a,. ;;::: 0 implies s,. ;;::: s,._1 , or, in other words, {s,.} is nondecreasing.
By the completeness property, such a sequence converges if and only if it is
bounded. I
Theorem 4 (A comparison test for convergencg) Let !r' b.. be a convergent

series. Let !r' a.. be another series, such that
0 ::::;:: a.. ::::;:: Mb.. (or all n (M = constant).
Then 2r' a.. also converges.
Proof The hypotheses clearly imply that the partial sums s .. = ~= 1 ak are
bounded and nondecreasing. Ht:nce !r' a .. converges. I
00 1 7T2
t It can be shown by methods of advanced calculus that ~
2n =- ::::::: 1.64.
1 6
Corollary (A comparison test/or divergence) Let !fen be a divergent series

of positive terms. Let !f an be another series such ~hat , .
an ~ men for all n (m = constant > 0).
Then ~fan diverges.
Theproofia
Example 6
!"" - 6n converges.
4
--
n=l n + 3 ---
bY. -
VI
L/
To see this, note that 6n/(n 4 + 3) < 6fn 3 Since the hyperharmonic series
!f l/n 3 converges, the given series mu~t also converge.
Another simple series which is often used for comparison is the following:
Theorem 5 (The geometric series)
In=o an= -1-a

1
- iflal < 1. (1.38)
I
The series diverges if Ia! > I.
Proof We know from Section 1.7, Formula (1.20), that
1 - an+l 1
sn = 1+a + a + + an =
2
-+ - - if Ia I < 1,
1-a 1-a
and this is the same as (1.38) above. Suppose on the other hand that lal ~ I.
Then limn-roan i=- 0, so that the series !~an must diverge. I
In using comparison tests, the following simple observation is useful.

Theorem 6 Let {an} and {b 0 } be two sequences o( positive numbers. JL

lim".... "' a 0 /bn exists, then there exists a constant M such that
an=::;;; Mb" for all n.
Proof Any convergent sequence is automatically bounded, so we know that
On =::;;; M = constant for all n. I JJ;tcff

bn II{;..)
,.' 'Wrf"'"
Corollary Let I an and L bn be series of positive terms.~ Assume that
limn-+"' anfbn ;;; L eXi$tS. Then
(a) if'J. b" converges, so does '2 an:

(b) if! bn diverges and L > 0, then .2 an diverges.
Proof Part (a) is obvious from Theorem 6. For (b), note that if L > 0, then
On> fL > 0
bn
for large n. Hence, I an must diverge if Ibn does. I
(\ P"' (~~~~~ { ,J ~
If a given series I an having all positive terms is encountered, it is fre-

quently (but not always!) possible to determine whether the series converges or
diverges by "common sense" reasoning, based on the theorems of this section.
First, if an does not approach zero as n ., oo, the series diverges But more
precisely, even if an _; 0 as n - oo, the series will diverge unless an - 0 "faster
than" 1/n. Conversely, if an- 0 "faster than" 1/n11 for some p > 1, then the
series must converge. Difficult cases arise when an approaches zero at approxi-
mately the same rate as 1/n, or when an behaves irregularly; see Exercise 3.
Example 7
The series
diverges, because the nth term behaves approximately like 1/3n (why?), and the
series I 1/3n is known to diverge. This can be proved rigorously by using the
corollary to Theorem 6.
We conclude by deriving one result concerning a special type of series having

both positive and negative terms, called an "alternating" series. By definition,~
series
~!
js called alternating provided that an> a;;:;;>.__O for all n,
n- oo. An example is the alternating-harmonic se\ ies
00 ( 1)"+1
I-n
1
=1-!+!-l+
Theorem 7 Every alternating series converges.
Proof Let s,. denote as usual the nth partial sum, ~~I, ( \ {) \ n , \
(i\ A/ 0 ' "'; -- VI} I
Note first that
o. ,,..?'~'~'>...l..o'r.e._ -::e,.~qr'e :") '.;,.,.~..,..)Er\;

It now follows from the completeness property that lim s2 ,. = S and
lim s2,._ 1 = S' both exist. Furthermore, (1.39) implies that
I
S2n :5: S :5: S' :5: S2n-1 <=) ')) (1.40)
S f-
, . ~ [":~." s~,. '
and therefore (given e > 0) __..- - - - ~
IS' - Sl ~ ls2n-1 - s2,.! ::0

(~___/p,
-f--t~ f --=j-
s r;' s
' ')? ~ ;t"- I
for sufficiently large n.
Consequently S' = S, and therefore also lim,._,. 00 s,. = S, or, in other
words, the series If (-1 )"+la,. converges to S. I
~
I-
--
From (1.40) it can be shownecfumte easilv,that ]
-f
~~i'l
e.~><
I'+'
S
!Jf#t) 1'1
~") J~"' /).,_..;:- )., - 'OihH
: l>ut A.= _(!~~~)
IL>,.. ,-.6,.1- G\ ,-, l.v...s+L 11 .f~
~-~~ ir-t-\:('r'JA 'f
52 Chopter One Sequences, Limits, and Real Numbers
in other words, the nth partial sum of an alternating series differs in absolute
value from the full sum, by at most an+ 1 ~~
s;:.
, ....
2~(--1 )"""'o.K I PI'\';;. L',.., [- ~ {""~ 1<. _::-; /J,_ = 1\-tt ( - \ ) D\~ s- z:-
~ 'f\ ~ PPh) -:) 'M:' ~ ~ S -4,., =- - 01"'~ + 'i;o.-tt-- Cl n -t-'l-+ '
I
t
r c..J
,,,.-J~
I c:. t
-a'~~"'
A- 01 -
~tl. 111 !'l-t'l-+
I Ifxercise'l
~ j-lil -1-0f \+ 1-V! +" It < "'
AJ....., ~t \ '1\t-'L '1\t'l ~ ~'f " 1/!i')-\1
~ fl\(~'f
1. , Decide whether or not each of the following series converges. Use the
corollary to Theorem 6 when convenient.
(a) i; n- 1, (b) ! n '

n=12n- 1 n=1 ,Jn 4 + 1
00 1
(c) ! '
n=1 ,Jn(n + 1)(n + 2)
oo n2 00 1
(e) ! -
n=12n
(cf. Section 1.10), (f) !
n=1 n
logn
2. Suppose that 2;:' a.. and!;:" bn both converge, and prove that
00 00 00
! 1
(ota,. + {3b,.) = oc 21 a .. + {3!1 b...
(You must use the definition of convergence, plus a certain theorem about
sequences.)
00 't
3. Show that 2 -n 1og- n diverges to + oo. Follow the method of Example 4;
n=2
you will be able to show that s .. > log [log (n + I)].
4. Let {s..} be a given sequence. Show that there is a uniquely determined

sequence {a..} such that {sn} is the sequence of partial sums of the series
If a... This shows (surprisingly) that the theories of sequences and of
series are logically equivalent! (Hint: s 1 = a1 ; s 2 = a 1 a 2 ; and so forth. +
Solve for a1 , a2 , )
5. Evaluate 2;' Ifn!.

oo I
6. Prove that 2
n=1 n(n + I) = I. (Try to guess a simple formula for s .. , and
then prove it.)
7. Prove the corollary to Theorem 4.

8. Criticize the following "proof" that 0 = 1.

co
(a) I ( -1)n = (1 - 1) + ( 1 - 1) + = 0 + 0 + = 0;
n=O
co
(b) I (-1)n = 1 + (-1 + 1) + (-1 + 1) +"
n=O
=1+0+0+=1.
Hence 0 = 1.
9. Prove by induction that
2
" 1 1 1 1 n
I-=1+-+-++->-.
k=l k 2 3 2n 2
Hence, the series I ~ can be shown to diverge without using integrals or

1
logarithms. n
*10. Discuss convergence of the series
1
1 - -2
2
+ -31 - -41 -f- -51 -
2
....
(This is not an alternating series!)

2 Limits, Continuity, and Differentiability
2.1 Introduction
The elementary study of limits of functions is strikingly similar to the

study of limits of sequences. The main difference lies in replacing "n-+ oo" by
"x-+ a" in appropriate places. As in Chapter 1, we will begin with some simple
numerical examples.
Example 1
It is "obvious" that lim.,_. 3 x 2 = 9. Determine a positive number <5 such that:
lx2 - 91 < 1
10 00 provided lx- 31 < <5.
Solution We write
lx2 - 91 = lx- 3llx + 31.
lx - 31 < 1, for example, then lx + 31 < 7; this can be

It is easy to see that if
seen from Figure 2.1 and can be derived easily from the triangle inequality.
See the lemma below.
54
2.1 Introduction 55
lx-3 1<1"
-3 0 3
Figure 2.1
Therefore, we have
lx2 - 91 = lx- 311x + 31

< 11x- 31 if lx- 31 < 1
<_1_ 1
if IX - 31 < - - .
1000 7000
Hence 15 = 1/7000 satisfies the requirement.
The following simple lemma is useful in calculations of "tolerances," d.
Lemma Jflx- al < h, then
(a) lx - bl < ib - al +h
and (2.1)
(b) lx- bl ~ ib -al-h.
The proof is immediate from the triangle inequality and is left to the reader
(see Exercise 1). Both inequalities of (2.1) are also quite obvious from an
examination of Figure 2.2.
,
xis here
l b-a l -h
a-h a b
~------~-------_J
l b-a l +h
Figure 2.2
56 Chapter Two Limits, Continuity, and Differentiability
Example 2
Consider the formula

lim x + 2 = - _! .
x-+-1 X - 3 4
Let e = I/IOOO. Find {J > 0 such that
lx + 11 = lx- (-1)1 < {J implies I:~~- (-~)I< e.

Solution First we see that
x+ 2_ (-.!)I = ~ lx + 11.
By (2.I)(b) we have
Ix- 3 4 41x- 31
lx- 31 ~ 13 + II - 1 = 3, provided lx + 11 ::;:: 1.

Therefore,
l xx+2+.!1=~1x+11
- 3 4lx- 4 31
1
<Six+ l iflx+11:::;:;1,
- 12 '
1 12 3
< lOOO, if lx + 11 < SOOO = 2A X 10-
Thus we can take {J = 2.4 x I0-3
In Examples I and 2 the values of the limits could be calculated simply by

substitution-for example, lim.,_ 3 x 2 = 32 = 9. This is not the case, however,
in many practical problems.
Example 3
Let
1/x
f(x) = 1
+ 1/x (x =F 0).
2.1 Introduction 57
Determine a= limx~of(x), and then determine b > 0 such that
1 -
lf(x)- a/ < 4 if /xi< b (x # 0).
Solution Of course we can simplify f(x) as follows:
1/x 1
f(x) = 1 + (1/x) - x-+-1 (x # O),
and now it is clear that limx~of(x) = 1. To find b, we have
lf(x)-11 = 1-+--11
x
1
1
=_lx_l_ (x#O)
/x + 11
lxl 1
<- if/xl<-,
1/2' 2
Therefore, we take b = min (1/2, 1/8) = 1/8.

The above example may seem artificial, but we have to begin with very
simple examples; less trivial ones will be encountered later. We consider finally
an example in which the limit does not exist, so that it is impossible to deter-
mine b.
Example 4
Let f (x) = xf lx l, and let L be a given real number. Show that no positive num-
ber b exists such that
lf(x) - Ll < 0.5 for all x with 0 < lxl < b.
(This shows that limx~of(x) # L for any L.)

58 Chapter Two Limits, Conti1111ity, and Differentiability
Solution Suppose that the inequality 1/(x) - Ll < 0.5 does hold for
0 < lxl < <5. Then, by the triangle inequality,
If(~) - f (- ~)I = k(~) - L + L - f (- ~)I
k - k(- ~) - I
~ (~) Ll + L
< 0.5 + 0.5 = 1.
Notice however (Figure 2.3) thatf(x) = 1 if x > 0, andf(x) = -1 if x < 0.
+It-------
------------,_------------~X
-------t-1
X
Figure 2.3 f(x) = lxl .
Therefore,
This contradicts our supposition.
Exercises
1. Prove the lemma of this section.
2. In each of the following problems we have lim.,_,.af(x) = L. Let e =

0.001. For each case determine some positive number <5 such that
lf(x) - Ll < e, provided lx - al < ~ and x =I= a.
(a) lim (x + 1) 3
= 1, (b) lim _1_ = 16'
a:-+0 o:-+1/4 x
2
2 +
33
(c) lim
.,_.5
(! - 2x) = -
X
49
5
, (d) lim x
z->2
2
X+ 2
+ 4 = 2,
(e) lim
1
=~
.,_._1 .j + 1
x2 ../2
113
(f) lim x = 2 (Hint: By algebra we have
.,_.s
x - 8 = (x 113 - 2)(x 213 + 2x 113 + 4).),
x2 - 1 1 -x-1
(g) lim-- = 2, (h) lim = -1.
.,_.1 X - 1 .,_.o 1 x-1+
3. Find the foUowing limits (if they exist) by inspection:
x 2 -1 1-x
(a) lim - 3- - (Factor!), (b) lim ,
.,_.1 X - 1 1
.,_.11-vx
xa 1
(c) lim-, (d) lim e-11"'
., .... o lxl ...... o
2.2 The Limit of 11 Function
In order for lim.,_af(x) to be meaningful, it is first necessary that the

function f(x) be defined for aU values of x in some neighborhood of x =a
(except possibly the point x = a itself). By a neighborhood of the point x = a
we mean an interval (a- p, a+ p) where p > 0. More precisely, the interval
(a - p, a pl is caUed thep-neighborhoodofx- a. We can now give the formal
"6-<5" definition of the limit of a function.
Definition Let f(x) be defined in some neighborhood of x = a, except pos-

sibly at the point x =a itself. Then lim'ha ((x) - L means that
for anY given 6 > 0 there is a corresponding !5 > 0 (which may depend
on 6) such that
lf(x)- Ll < 6 (or all x satisfying lx- al < <5, x =I= a. (2.2)

Notice in the definition that we do not assume thatf(a) is defined (but it
may be!); therefore, in (2.2) the point x =a is excluded from consideration.
This is an important observation, particularly when it comes time to define
derivatives in terms of limits. Also we do not suppose that 6 (or <5) is "small."
To do so would complicate the definition unnecessarily, and anyway we would
then have to define "small." LogicaUy speaking, if Condition (2.2) holds for aU
"smaU" 6 > 0, it automaticaUy holds for all 6 > 0, and conversely. Of course
in practice it is usually the "small" values of 8 which have to be especially taken

into account.
If Condition (2.2) is satisfied, we say that the limit off(x) as x approaches a
exists and equals L. (Naturally Lis unique: see Theorem 1 below).
Notice how similar (logically) this definition is to the definition of the limit
of a sequence. We might say that the two definitions are the "same" except for
certain necessary "adjustments." This similarity means also that the elementary
theory of limits of functions closely resembles that of sequences.
We illustrate the procedure involved by a few examples.
Example 1
Show that 1im.,_.3 x 2 = 9.t
Solution Let e > 0. Then

lx2 - 91 = lx + 311x - 31
< 7 lx - 31 (if lx - 31 < 1)
<8 (if also lx - 31 < 8/7).
Let IJ =min (1, 8/7). Then we have shown that
lx2 - 91 < e, provided lx - 31 < IJ.

Note that the simple choice IJ = 8/7 does not work for every 8 > 0.
Example 1 is quite trivial, at least in the sense of evaluaimg lim.,_ 3 x 2
We will therefore give some less obvious examples. The reader would profit by
trying to find the limits in question-for example, by examining the graph of
f(x)-before reading the solutions.
Example 2
xz
Find lim - I , and prove it.
:c-o 1x
Solution Notice that lx 2/lxll = lxl if x =1= 0, so it should be obvious that

the limit is 0 (See Figure 2.4). In fact, if 8 > 0 is given, we can take IJ = 8;
then
2
x 1 = IxI < 8 if IxI < IJ = e and x =I= 0.

I lxl
t You may have noticed that in many of our examples we have lim- f(x) equal to [(a).
As we shall see in Section 2.4, this situation is characteristic of continuous functions f(x).
Figure 2.4 f(x) = x 2/lx!.t
Example 3
Define a function g(x) for -1 ~ x ~ I as follows:
1 ifxisanyofthenumbersl,.!,!, ... ,.!, ... ,

()
gx = . 2 3 n
{
otherwise.
0
(See Figure 2.5). Find
(a) lim g(x); (b) lim g(x); (c) limg(x).
z-+3/8 z-+1/4 z-+0
Prove your answers.
X
-I \14 3/s
Figure 2.5 The function g(x) of Example 3
t We use the convention, in sketching graphs, that a small circle on the graph indicates
that the point encircled is not on the graph of f(x).
62 Clulpter Two Limits, Continuity, and DifferentlabUity
Solution (a) lima:--.818 g(x) = 0. To prove this, let e > 0 be given. Choose
15=1-!=-h
(in this case 15 does not depend one!). Then the interval ((3/8) - 15, (3/8) + 15)
contains none of the numbers 1, 1/2, 1/3, ... , so that
lg(x)l = 0 < e for all x with lx - II < 15.

This shows that g(x) -. 0 as x-. 3/8.
(b) limz_ 11,g(x) = 0. To see this, we take
(again independent of e). Then for any given e > 0,
lg(x)l = 0 < e if lx - !I < 15 and x = t

(Note that g(1/4) = 0, but this does not affect the value of lim-114 g(x).)
(c) lima:--.0 g(x) does not exist. To show this, suppose we have
lima:-o g(x) = L. Take e = 1/2; then there must exist 15 > 0 such that
lg(x) - Ll < t whenever 0 < lxl < 15.

But we can select two points x 1 and x 2 such that 0 < lxil < <5 (i = I, 2), and
g(x1) = 0 and g(x2 ) = 1. Therefore we have
lg(xt)- Ll =ILl< t and lg(xa)- Ll = 11- Ll < t.

Since both these inequalities cannot be satisfied by any real number L, we
conclude that L does not exist.
We proceed next to the most elementary theorems about limits of func-
tions. These theorems, as well as their proofs, are closely analogous to theorems
proved in Chapter 1 for limits of sequences.
Theorem 1 /[lim.....,. f(x) exists, it is unique.
Proof Suppose 4 and L 2 are both limits of f(x) as x-. a. Then, given any
e > 0, there must exist a number <5 > 0 such that both
lf(x) - Ltl < ! and 1/(x) - Lsi <!

2 2
provided 0 < lx - al < 15. But then
Since e is arbitrary, we conclude that L 1 = L 2 I
Theorem 2 Suppose that
limf(x) =L and lim g(x) = M.

P:=t"
Then
(i) lim [/(x)

....
+ g(x)] = L + M,
(ii) limf(x)g(x) = LM,

a:-+
(iii) lim f(x) = L if M "# 0.

z-+ag{x) M
We leave the proofs of (i) and (ii) as exercises. To prove (iii) we first estab-
lish the following lemma.
Lemma Suppose that lim.,.....a g(x) = M "# 0. Then there is some number
do > 0 such that
lg(x)l > ! IMI for 0 < lx - al < d0
Proof If d0 > 0 is chosen so that
lg(x) - Ml < ! IMI for 0 < lx - al < d0 ,

th~n we have
lg(x)l = IM - (M - g(x))l
::=:: IMI - IM- g(x)l
> IMI - ! IMI = ! IMI for 0 < lx - al < d0 I

6tl Chapter Two Limits, Continuity, and Differentiability
Proof of Theorem 2(iii) First consider the casef(x) - It. We have
1 1 I= lg(x) - Ml
Ig(x)- M IMIIg(x)l
21g(x)- Ml
< M2 if 0 < Ix - aI < <5 0 ,
where <5 0 > 0 is determined by the lemma. If e > 0, let <5 1 > 0 be chosen so that
eM 2
lg(x)- Ml < - for 0 < lx- al <<51
2 . .
Finally, let b = min (<5 0 , <51 ); then we have
1
- - - _!_ I < e for 0 < Ix - aI < o.
1 g(x) M
This shows that lim.,~a 1/g(x) = 1/M.

To prove (iii) for an arbitrary numerator f(x), we assume that part (ii)
has already been proved. Then we know that
1 1
limf(x) = limf(x) - - = limf(x) lim - - =.f._ . I
x-+a g(x) x-+a g(x) x-+a x-+a g(x) M
Theorem 1 Given f(x) < g(x) [or every x in some neighborhood of x = a,

suppose that
limj(x) = L and lim g(x) = M.
ThenL < M.
The proof, similar to the case of sequential limits, is left as an exercise.
Note that ifj(x) < g(x) for every x in some neighborhood of x =a, we cannot
necessarily conclude that L < M.
Exercises
1. Give a verbal description of the definition oflim.,~af(x) = L, to accompany

Figure 2.6.
t The notation f(x) == 1 (''[(x) is identically equal to 1") means that f(x) = 1 for all x.
2.2 Tlie Limit of a Function 65
The point (a, L)

An F-neighborhood
of L-......,.._
L __ _ -~~he graph of f(x)
--~--------~~-------------+X
a "'A 8-neighborhood of a
Figure 2.6 limf(x) = L.

x~a
2. Prove the following "obvious" limit formulas by the e-15 method (cf.
Exercise 1 of Section 2.1).
2
(a) lim (x + 1)3 = 1, (b) lim x +4= 2,
x-+0 x-+2 X+ 2
= ~- .
2
1
(c) lim
x-+-I../x 2 +1 y
3. Make a reasonable sketch of the graph of f(x) = /~j on the interval

-1 :::;; x:::;; I. What is lim.,~ 0 f(x)? Give the proof.
1
4. Show by the e-15 method that lim --- = 0. (The inequalities needed may
, _o 1og lx 1
be confusing. Remember that log lxl < 0 if 0 < lxl < 1.)
5. Give an e-15 proof of the fact that lim ~ does not exist.
x~ox
6. Prove Theorem 2(i).
7. (a) Define "bounded" for a functionf(x) defined on a given interval/.

(b) Prove the lemma: If lim.,~af(x) = L, then f(x) is bounded on some
~5 0 -neighborhood of a.
8. Prove Theorem 2(ii). (Use the lemma of Exercise 7.)
9. Suppose that lim.,~nf(x) = L.

(a) Prove that lim.,~a lf(x)l = ILl.
(b) Prove that if L > 0, then lim.,~a ..}f(x) = /i.
10. Prove or disprove the following statements.
(a) limf(x) = 2limf(x), (b) Jimf(2x) = 2limf(x),
(c) limf(x) = Iimf(2x).

::a:-+2a z-+a
11. Give the proof of Theorem 3.
2.3 Other Types of Limits
Sometimes one wishes to find the limit of a given function f(x) as x ap-
proaches a from one side. For example,letf(x) = e-1 '"' Then, as we show below,
f(x)-+ 0 as x-+ 0 with x > 0, whereas lim.,....0 f(x) does not exist. See Figure
2.7.
y
_L_
Figure 2.7 f(x) = e-lla:;
The formal definition of one-sided limits is obtained by making the obvious

adjustment to the basic definition of Section 2.2. Namely, we say that
Iimf(x) =L
11:--+a+
_jf the following condition holds:
for any given e > 0 there is a number c5 > 0 such that
lf(x) - Ll < e whenever a <x< a + 5.

Example 1
lim..,_. 0 + e-1 f:e = 0. To prove this, let e > 0 be given. If e < 1, then the
required inequality
can be solved as follows :t
log e-11== _l <loge;

X
equivalently,
x loge> -1 (since x > 0 by assumption);
that is,
1
x < - - - (since loge < 0 for e < 1).
loge
Hence, if we take
1
15 = - - - when e < 1
loge
and
15 = anything when e ~ 1,
then we have
e-1/% < e if 0 < x < 15.
(Why does this hold when e ~ 1 ?)

Next we consider limits that involve infinity. There are several possibilities,
since we may have x - + oo or - oo, or f (x) - + oo or - oo. In every case the
formal definition is obtained by making a suitable adjustment in the basic limit
definition. We will give one example, and ask the student to give other definitions
himself (see the Exercises).
Definition Let f(x) be a function defined for all x > x 0 for some x 0 We
write
limf(x) = L
:e-++ao
j[_the following condition is satisfied:

given any e > 0 there exists a real number M such that
lf(x)- Ll < e whenever x > M .
t We use here the fact that log x is an increasing function of x.

68 Clulpter Two Limits, Contilfllity, and DijJerentiability
Example 2
Find M such that

11 < e
l _x_ -
x+2
for all x > M.
(This will prove that lim _!._ = 1.)

z-+ooX + 2
Solution This is easy.
_x_ - 11 = _2_ < ~ < e,

l x+2 x+2 x
provided x >~= M.
e
Example 3
lim.,._.. 0- e-1 /:x: = + oo (see Figure 2. 7).

We can check this as follows:
lim e- 1 /:x: = lim e-t (replacing x by 1ft)

z-+0- t-+-co
=lim eu = +oo.
u-++c:o
The method of replacing x by 1/t in this example is easily justified by

referripg to the definitions. We give the following theorem as one example of
this process.
Theorem I Let f(x) be defined on an interval 0 < x < a. JLeither of the

limits.
lim f(x)
:x:-+0+
or lim f
t-++oo t
(!)
exists, or if either limit is + oo or - oo, then both limits have the same value.
Proof Suppose for example that Iim:x:-o+f(x) =L (L =f- oo). If e > 0 is

given, there exists t5 > 0 such that
lf(x) - Ll < e if 0 < x < b.

Write x = 1ft; then
This shows that limt~+oo/(1/t) = L. A similar argument applies if L is + oo or

-00 .
To write out a complete proof of the next theorem would be most tedious.
By now you should be confident that you could prove any part of this theorem if
necessary.
Theorem 2 Let limf(x) denote any one of the limits
lim f(x), lim f(x), lim f(x) or lim f(x),

x-+a+ x~a- ::t-t-+oo x-+-oo
and assume lim ((x) and lim g(x) exist and are finite. Then, under the usual
hypotheses, all algebraic and order properties are preserved by taking this limit.
Exercises
I. Find the following limits by inspection. Omit proofs. (Some limits may be
infinite.)
(a) limlogjxj, (b) lim Ilog xl,

o:-+0 o:-+0+
(c) lim _x_, (d) lim tan x,

o:-+-oo 1 - 2x o:-+~r/2-
(f) lim sin x '

o:-++oo X
1
(g) lim arc tan x, (h) lim - - .
x-++oo o:-+2- X - 2
ella:
2. Letf(x) = (x #- 0). Find
1 + e11"'
(a) lim f(x), (b) lim f(x),
x-+0+ o:-+o-
(c) lim f(x), (d) lim f(x) .
(Use Examples I and 3 of the text.) Sketch the graph of f(x).

3. Write down the formal definition for each of the following limits. L and a
always denote finite numbers.
(a) lim f(x) = L, (b) limf(x) = +oo,

a::-+a- z-+a
(c) limf(x) = -oo, (d) lim f(x) = +oo,

z-+a o:-+a+
(e) lim f(x) = + oo.

o:-++oo
4. Show that lim.,__.+oof(x) = limt--oof( -t) if either of these exists.
5. Prove that if lim.,__.af(x) = +oo and lim.,__.ag(x) =L (finite), then
lim [f(x) + g(x)] = + oo.

z-+a
(The following proof, though short, is unacceptable:
lim [f(x) + g(x)] = limf(x) +lim g(x) = +oo + L = +oo.)

z-+a z-+a z-+a
6. Give the complete proof of the statement: Iff (x) ::;; g(x) for all x, then
lim f(x) ::;; lim g(x),

:e-++a::~ o:-++co
assuming these limits exist and are finite. Do the limits have to be finite?
7. What can be concluded if one knows that lim.,__.a+f(x) = lim.,__.a_f(x)?
8. Is it true or false that in every case
lim f(x) = lim f(n)?

Q:-++oo n-++oo
(This may be controversial!)
9. Criticize the following calculation:
lim
z-++oo i "'
0 X
1
- dt = iQ() 0 dt = 0.
0
2.4 Conti1111ity 71
2.4 Continuity
From early times man has pondered the question of the continuous.
Mathematicians still refer to the real-number system as the "continuum";
physicists talk about the "space-time continuum." The question of the con-
tinuous versus the discrete nature of matter was hotly argued by scientists and
philosophers for centuries. These arguments carried over into pure mathe-
matics, and led to such confusing notions as "infinitesimal," "instantaneous,"
and so on. One of the principal causes of confusion was the failure to separate the
mathematical abstractions from physical reality. A completely satisfactory
definition and theory of continuity was finally developed by nineteenth-century
mathematicians, particularly K. Weierstrass (1815-1897) and J. Dedekind
(1831-1916).
The first step in reaching the definition seems to have been the realization
that continuity should refer always to functions 1 rather than to vaguely defined
objects such as curves or surfaces. (In fact, nowadays, curves and surfaces are
themselves defined in terms of continuous functions.) Most people have an
intuitive idea of what a continuous graph is-a graph is continuous if it is
"unbroken." As in Figure 2.8, this property seems to refer to the whole graph.
y y
~I
I
I
Continuous graph Graph with discontinuities
Figure 2.8
The next step in understanding continuity is the realization that if a graph is

broken at a single point, it is discontinuous (at that point, at least!), but other-
wise it is continuous.t
Continuity at a given point x =a, however, can immediately be defined
in terms of the limit as x-+ a:
t It is interesting to note that the Russian word for continuous is nye'prerivni, which
means, literally, un'broken.
Definition I Let f(x) be a function defined in some neighborhood of x = a.

Then ((x) is s'Jid to be continuous at x - a pro11ided that
limf(x) = f(a).
x->a
The function f(x) is said to be continuous an a given interval (oc, {1) if it is con-
tinuous at each paint of this interval
Most functions of elementary calculus are continuous. For example, we

can immediately prove the following theorem.
Theorem I Every polynomial function p(.x) is continuous at every poinJ.
Proof First, it is completely trivial to prove that
lim x =a.
Now let p(x) = c0xn + c1xn-1 + + en be a given polynomial. Then by the

basic theorem (Section 2.3) about sums and products, we conclude that
lim p(x) =lim (c 0 xn

x~a x-ta
+ c1xn-1 + +en)
This proves that p(x) is continuous at x = a. I
Example I
Let f(x) = lxl (Figure 2.9). Since lim.,~a lxl = lal for any number a (even
including a= 0), it follows that f(x) is continuous everywhere.t
t However, note that f(x) = Ixi is not smooth at x = 0; that is, it does not have a deriva-
tive at x = 0.
2.4 Continuity 73
Figure 2.9 f(x) = lxl-
A function which is not continuous at a I?oint x = a is said to have a

discontinuity at that point There are several possible kinds of discontinuities,
and the next several examples illustrate progressively "worse" cases.
Example 2
(A "trivial," or "removable," discontinuity.) Let
f(x) = ( + ~1)-1
x- 1 1 (x op 0, - 1),
as in Figure 2.10. Since f(O) is not defined, the definition of continuity at x = 0

cannot be satisfied. But we have lim.,_ 0 f(x) = 1. We could therefore "extend"
the definition of f(x) by defining f(O) = 1. The extended function f(x) is
continuous at x = 0 (but not at x = -1).
Figure 2.10 f(x) = .x- 1 1 ( + x1)-1.

74 Cluzpter Two Limits, Continuity, and Differentiability
Definition 2 A function f(x) defined for all values of x in some neighborhood

of x = a, except possibly for x = a itself, is said to have a trivial or removable
discontinuity at x = a i[
Iimf(x) = L
o:-+a
exists, but L =I= f(a). In this case the definition off(x) can be modified to obtain
a function continuous at x = a, by (re)definingf(a) = L.
Example 3
Letf(x) = sin x (x =1= 0). It is proved in elementary calculus texts that

X
. sin x
Itm-- = 1.
a:-+0 X
Hence,f(x) has a removable discontinuity at x = 0 (see Figure 2.11).
1:1gure
E" 211 f(x) -- sin x .
X
Example 4
Letf(x) = l:l (x =I= 0). It is easy to see (Figure 2.12) that
lim f(x) = +1 and lim f(x) = -1.

a:-+0+ a:-+o-
This is an example of a function having a "jump discontinuity" at x = 0.

2.4 Continuity 75
X
Figure 2.12 f(x) = lxl
Pefinitiaa I A function f(x) defined in a neighborhood of x = a, except

possibly for x =a itself, is said to have a jump discontinuity at x =a iflim.,.....a+
f(x) and Iim.,.....a- f(x) exist (finitely), but are not equal. The value
Ja(f) =lim f(x)- lim f(x)

a:~a+ a:-+a-
is called the iump of f at a.

' Another example of a jump discontinuity is provided by the function (see
Exercise 2, Sectiori 2.3)
et'"'
f(x) = 1 + et'"' .
Example 5
Let f(x) = sin 1/x (x =I= 0). This funtion has an oscillating discontinuity at
x = 0. In this case the oscillations are bounded.
Figure 2.13 f(x) = sin !X .

Example 6
The functions shown in Figure 2.14 illustrate various kinds of infinite discon-
tinuities at x = 0. The reader can provide formal definitions of these sorts of
discontinuities if desired.
In all of the above examples, the functions f(x) have only finitely many
points of discontinuity. We next give two much more complicated examples,
in which there are infinitely many points of discontinuity in a finite interval.
y y y
f(x) =xI f(x) = I

x' f(x) = e7'
Figure 2.14
Example 7
Define a functionf(x) as follows:

if x is rational,
j(x) = {~ if x is irrational.
Since any interval (a- e, a+ e), e > 0, contains points at which f(x) = 0
and points at whichf(x) = l, it is clear that lim.,_af(x) does not exist. Hence
f(x) is not continuous at any point x = a.
Example 8
(Dirichlet's function). Define a functionf(x) on 0 < x ::;; l by setting
.! if x = !!. in lowest terms,

f(x) = q q
{
0 if x is irrational.
1.4 Continuity 77
Some examples of values ofj(x) are :
/(1) = 1, j(t) = t, /(1 00) =

99
1~0, 1(p~2) = 0,
and so on. It is easy to see that the total number of points x, 0 < x < 1, at
whichf(x);;;::: 1/q is at most (1 /2)q(q- 1). See Figure 2.15. Let us show that for
y
. . . .
"..... . =. .... .. :: ... : ...,
----~------~~--~----~----~x
0
Figure 2.15 Dirichlet's function.
the Dirichlet function we have
lim.,-.af(x) =0 (2.3)
for every a. Since f(a) = 0 if and only if a is irrational, this will prove that
. . irrational number and discontinuous at
Dirichlet's function is continuous at everv
.
everv rational number .
To prove (2.3), let e > 0 be given . Choose a positive integer q such that
1/q < e. Then we can find some positive number~ such that the interval (a -- ~.
a+~) contains none of the finitely many points x at whichf(x);;;::: 1/q, except
possibly the point a itself. (Can you show how to determine~?) This means that
1
lf(x)l = f(x) <- < e
q
for all x with lx- al < ~. x -:f= a, since for such values of x, eitherf(x) = 0 or
f(x )= 1/r with r > q. This completes the proof of (2.3). ~
Let us return to the study of functions that are continuous. Since continuity
is defined in terms of limits, and since sums, products, and quotients are pre-
served by the limit operation, we obtain the next result.
Theorem 2 Suppose that f(x) and g(x) are continuous at x =a. Then the
functions
(i) f(x) + g(x),

(ii) f(x)g(x),
(iii) f(x)fg(x) (if g(a) =I= 0)
are also continuous at x ,;:;, a.
Proof To prove (i), for example, we have
lim [J(x) + g(x)] = limf(x)

., ... a
+ lim g(x)
= f(a) + g(a),
by hypothesis. This proves thatf(x) + g(x) is continuous at x =a. The proofs

of (ii) and (iii) are similar. I
Next we consider the case of a composite functionf(g(x)) .
Theorem 3 Let g(x) be continuous at x = a, and f(t) continuous at t = g(a).

Then the composite functi~n f(g(x)) is continuous at x =a.
The functionj(g(x)) is called the composite (unction of [and g; Theorem 3

says that the composite o(.llY.o_c_<mtinl!..~1io.ntis..a.gainJ! continuous func-
tion.
Proof We have to show that
limf(g(x)) = f(g(a)) . (2.4)

., ... a
Let e > 0 be given. Since f(t) is continuous at t = g(a), there exists <51 > 0
such that
lf(t) - f(g(a))l < e if It - g(a)l < b1. (2.5)
e.;
2.4 Continuity 79
But since g(x) is continuous at x = a, there exists <5 2 > 0 such that if lx - al <
<5 2 , then
lg(x) - g(a)l < <51.
Hence by (2.5) with t = g(x),

lf(g(x)) - f(g(a))l < e if lx - al < <5 2
This proves (2.4). I
Example 9
Letf(x) = "'l(x- I)(x- 2)1 (see Figure 2.16). We know that
(a) (x - I)(x - 2) is continuous for all x (Theorem I);

(b) ltl is cbntinuous for all t (Example 1); and
(c) "'~ is continuous for all u ~ 0 (Theorem 4 below).
Hence, applying Theorem 3 twice, we conclude thatf(x) is continuous for
all x.
y
Figure 2.16 f(x) = "'l(x - l)(x - 2)1.
Theorem 4 Thefollowingfunctions are continuous:
_ill_ sin x and cos x,for all x;

.ill) eX ,for all x;
..(iii.).. log x,for all x > 0;
(iv) xP (where pis any real number),for all x > 0.
80 Chapter Two Limits, Contillllity, and Differentiability
The proof of Theorem 4 will not be given here. Later, in Chapter 4, we will
define the exponential, logarithmic, and trigonometric functions carefully and
prove that they are continuous. Meanwhile, although we will not hesitate to
use these functions in examples and exercises, we will avoid using them in the
logical development of the theory, thereby avoiding the danger of logical
"circularity."
Exercises
1. Each of the following expressions defines a function f(x) for all x E fJI
with finitely many exceptions. Determine all these exceptional points and
investigate the type of discontinuity thatf(x) possesses at each exceptional
point. For removable discontinuities, calculate lim.,_af(x). For jumps,
calculate Ja(f).
1 x-1
(a) lxl- 1' (b) lxl- 1'
x-1 (d) __1-,

(c) lx- 11' smx
.) sin x (f) x sin!,
(
e .J1x1' X
(g) [lOx] ([ ] =greatest integer), (h) log I~ j.

(i) log\xl'
(j) .! sin log lxl .
X
2. Let '(T) denote the amount of heat which must be applied to raise a given
block of ice at 0F to a temperature of TF. For which values ofT is '(T)
discontinuous, and what is the physical significance of the discontinuity?
3. Dirichlet's function (Example 8) is discontinuous at x = 1/2. What type of

discontinuity is this?
4. Sketch carefully the-graph ofj(x) = x 2 ' 3 near the origin. Does this function
appear to be continuous at the origin? Is it?
5. Let 0 < x1 < x2 < < xn < I be finitely many points. Define functions
(fork= I, 2, ... , n). Sketch the graph of
and discuss its continuity.
6. Ifj(x) is a function defined for a :::;:; x :::;:; a+ {J (some lJ> 0), we say that
f(x) is right-continuous at x = a if
lim f(x) = f(a).

:z:-t-a+
Discuss in terms of right continuity
(a) [x],
(b) the function of Exercise 5.
7. Give an example of a functionf(x) defined for all real x, but continuous

only at x = 0.
8. Give an example of a functionf(x) which is continuous nowhere, but with

(j(x)) 2 continuous everywhere.
9. Define f(x) by the formula

x"
f(x) =lim - - 11.
I n->co 1 + X
Calculate j(x) for all possible x and discuss its continuity.
2.5 Differentiability
Historically the notion of the derivative of a function preceded by about

two centuries the notion of continuity of a function. This observation is re-
flected in the relative ease with which students learn about derivatives, in
comparison to the difficulty they encounter in understanding continuity.
In this section we review the definition of derivative and prove that a
function ((x) is differentiable at a given point x = a if and only if the graph of
,f(x) "looks arbitrarily like a straight line" in sufficiently small neighborhoods of
82 Chapter Two Limits, Contillllity, and Differentiability
y y
-1 -1
f(x)= x' g(x) = JxL
Figure 2.17
~ This very important characterization of differentiability is seldom

stressed in calculus books.
To illustrate this idea of differentiability, we consider the following two
simple examples (Figure 2.17):
(a) f(x) = x 2 ,
(b) g(x) = lxl.
We know that both f(x) and g(x) are continuous at x = 0. To see how the
graphs look in a "small" neighborhood of x = 0, let. us magnify each graph
ten times; see Figure 2.18. Note that "n~ar" x = 0, the graph of f(x) is
y y
- 1 1 - 1 1
lo lo lo lo
f(x) =x' g(x) = lxl
Figure 2.18
approximately a straight line, but g(x) has the same sharp corner at x = 0, no
matter how close one is to x = 0. In factf(x) = x 2 is differentiable at x = 0,
but g(x) = lxl is not. Differentiable functions are sometimes said t> have
"smooth" graphs. We now transpose the foregoing description into mathe- natical
terms.
Definition 1 Let f(x) be defined in a neighborhood ofx =a. Ihe derivative of

[(x) at x =a is defined as
'( a) -1m
_ 1. f(x) - f(a)
! x->a X- a
,
provided this limit exists lf the limit does exist, we say that f (x) is differentiable
at x =a.
Theorem 1 1/f(x) is differentiable at x = a, then [(x) is continuous at x = fl
. . f(x) - f(a) . . f(x) - f(a)

Proof Smce hm exists, the functiOn must be bounded
.,....a x - a x-a
near x =a:
f(a) I ~ M = const
lf(x)-
x-a
(0 < lx- al <!5o)
If e > Q is given, let !51 = efM. Then if 0 < lx- al < <5 =min (!5 0 , !51 ),
we have
lf(x) - f(a)l ~ M lx - al < e.
This proves that lim.,.....af(x) = f(a). I
Definition 2 Let f(x) be defined in a neighborhood of x = 0. We say that

f(x) is of smaller order than x as x ~ 0 provided that:
for any given e > 0 there exists <5 > 0 such that
lf(x)l ~ e lxl whenever lxl < <5. (2.6)
Example 1
Letf(x) = x 2 Then
lf(x)l = lx2 1= lxl lxl ~ e lxl
provided that lxl < e. Hence, we may choose <5 = e in order to satisfy (2.6),
so we conclude that x 2 is of smaller order than x as x ~ 0.
Example 2
Letf(x) = lxl, and suppose 0 < e< 1. Then the inequality
lf(x)l = lxl ~ e lxl
cannot hold except for x = 0. This means that lxl is not of smaller order than
x as x~o.
Graphically, Condition (2.6) simply means that the curve y = j(x) lies
between the lines y = ex when lxl < !5 (see Figure 2.19). If [(x) is a function
of smaller order than x as x ~ 0, we say that the curve y - f(x) is tangent to
the x-axis at the origin. This definition of tangent can be generalized easily as
follows.
Figure 2.19 lf(x)l < e lxl for lxl < !5.
Definition 3 Let j(x) be defined in a neighborhood of x =a. The curve

y= f(x) is said to be tangent to the straight line
y = m(x- a)+ b (2.7)
at x = a, provided that j(x) - m(x - a) - b is of smaller order than x - a as

x ~a. In other words, the curve y = f(x) is tangent to the line (2.7) if, given
e > 0, there exists !5 > 0 such that
lf(x) - m(x - a) - bl ::;; e lx - al when lx - al < !5. (2.8)
Thus we see that the definition of tangent requires an e-!5 statement.t
t Definition 3 can be generalized in a very simple and obvious way to define tangent
plan_!:s to surfaces z = f(x, y), and even to more general situations in n dimensions.
Theorem 2 Let f(x) be defined in a neighborhood of x =a. Then the curve

y = f(x) is tangent to the line y = m(x- a)+ bat x =a if and only iff(x) is
differentiable at x =a andf(a) = b andf'(a) = m.
Proof First assume y = f(x) is tangent toy= m(x- a)+ b at x =a. If we

put x = a in (2.8), we get
lf(a) - bl ~ 0,
so thatf(a) =b. Using the inequality (2.8) once again, we now have, for each
e > 0, a ~ > 0 such that
lf(x)- m(x- a)- f(a)l

lx- al
= lf(x)- f(a) _ m
x- a
I< e
when 0 < lx - al < <5. But this means precisely that
limf(x)- f(a) = m.
o:-+a X- a
Thusf(x) is differentiable at x =a andf'(a) = m.

The converse is proved simply by tracing backward the steps of the above
proof. I
Example 3
The Lions Gate bridge at Vancouver, B.C. (Plate 1) provides an interesting

illustration of nondifferentiability. The bridge is famous partly because of a
sudden "bump" in the roadway at the center of the span. In fact, the curve of
the roadway is nondifferentiable at this point. The sudden change in the slope
of the roadway is quite evident to anyone driving over the bridge. (You may
wish to ponder to what extent this description is a mathematical abstraction
of the actual physical condition of the bridge.)
Suppose next thatf(x) is defined on an open interval (a, b) and thatf'(x)

exists for each x in (a, b). Thenf'(x) is a function on (a, b), and we can consider
whether f'(x) is continuous or differentiable, and so on.
Definition 4 Let f(x) be defined on an open interval (a, b). We say that ill)_
is continuously differentiable on (a, b) i(f'(x) is defined and continuous on (a, b).
Plate 1 (Courtesy of The Vancouver Public Library, Historic Photograph Section.)
y y
f(x) f'(x)
(a) (b)
Figure 2.20
More generally, we say that [(x) is k times continuously differentiable on (a, b) if
exists and iv continuous on (a, h).
Example 4
The functionf(x) defined by

x2 if X~ 0,
f(x) = {0 ifx < 0
is continuously differentiable, but not twice continuously differentiable, on

( -oo, oo).
To see this, we notice that
2x if X~ 0,
f'(x) =
{0 if X< 0,
so that f' (x) is a continuous function, but not a differentiable function. See
Figure 2.20.
Example 5
As an application of continuous differentiability we consider the problem of

designing highway curves. Let the curve y = f(x) represent the center line of a
(level) highway. It is certain thatf(x) must be continuous, and the most elemen-
tary dynamical considerations show that f(x) must also be differentiable-
see Figure 2.2l(a). A little more reflection on the dynamics of an automobile
leads to the conclusion that the curvef(x) should be at least twice differentiable.
Highway engineers construct twice differentiable curves by fitting t.:>gether
cubic curves (j(x) = ax 3 + bx2 +ex+ d) . They may be unaware of the
possibility of designing infinitely differentiable highways by utilizing the principle
of the following example.
A continuous highway
~
A differentiable highway
(a) (b)
Figure 2.21 Inadequate highway design.
Example 6
Letf(x) be defined by
e-l/z, if X > 0,
f(x) = {0, if X :o;;: 0.
Then f(x) is infinitely differentiable for all x; that is, j<k>(x) exists for every
positive integer k.
To see this, note first that, from calculus,J<k>(x) exists for any kif x ;;/= 0.
We can show that
j<k>(O) = 0 (k = 1, 2, 3, .. .),
and this will prove the assertion. Fork = 1 we have
f'(O) = limf(x)- f(O) = limf(x).

Z->0 X Z->0 X
Since f(x) = 0 for x < 0, obviously lim., __ 0_f(x)fx = 0. For x ~ O+ we

have
lim f(x) =lim ! e-11"' = lim te-t = 0,

z->0+ X z->0+ X t->+oo
by a well-known property of exponentials. Hence f' (0) = 0. Similarly
f"(O) = lim f'(x) - f'(O) = lim f'(x) .

., .... o X "'-+0 X
The limit from the left is again obviously zero; since f' (x) = x- 2e-1 '"' for x > 0,
we find that
Therefore, f" (0) = 0. It is fairly easy to extend this argument, and prove by
induction thatj<k>(O) = 0 for all k.
The function of Example 6 may seem an idle curiosity. But the existence of
such functions is an absolutely essential ingredient of a vast amount of con-
temporary mathematical analysis, including mathematical physics.t
Exercises
I. Determine which of the following functions are of smaller order than x

as x --+ 0. Give reasons.
2
(c) sin 2 x, (d) _x_.
x+1
2. Each of the following functions is continuous everywhere. Determine all

points at which each function is nondifferentiable. (A sketch of the graph
will be quite revealing.)
1
(a) lxl +1;
(c) !sin xi; (d) lxl + lx- 21.
3. Give an example of a function which is k, but not k + I, times continuously
differentiable.
t See LSchwartz, Theorie des distributions, Hermann (Paris) (1966).

4. The right-hand derivative ofj(x) at x =a is defined to be
' ( ) _ 1. f(x) - f(a) .

! +a -1m x-+a+ X- a
Find the right- and left-hand derivatives of each of the functions (a)-(d)
of Exercise 2, at each of its points of nondifferentiability.
5. Explain geometrically the meaning of tangent line given in Definition 3

(cf. Figure 2.19).
6. Explain why the curve of Figure 2.21 (b) is not suitable for a high-speed
highway. (How would you steer an automobile through such a curve?)
7. Complete the proof by induction, that f'kl(O) = 0 for all k, where f(x)
is the function of Example 6.
*8. Find a functionf(x) on ( - oo, oo) satisfying the following conditions (see
Figure 2.22).
Figure 2.22
(a) j(x) is infinitely differentiable,

(b) f(x) ~ 0 for all x,
(c) f(x) = 0 if lxl ~ 1,
(d) f(O) = 1.
*9. Do there exist differentiable functions which are not continuously differ-
entiable? Explain.
3 Properties of Continuous Functions
3.1 Introduction
In Chapter 1 we stressed the importance of the completeness property of

the real-number system &l in studying the convergence of sequences and series.
In this chapter we show how the completeness property is used to prove certain
basic theorems about continuous functions. For this purpose, -and for- -many
other applications in analysis, it is necessary to consider various other properties
of &l which are closely related to completeness. Such properties are discussed in
Sections 3.2, 3.3, and 3.4.
In the remaining sections we prove some b~.sic theorems about continuous
functions, such as the theorem that a continuous function defined on a closed
bounded interval, a :<:;;: x :<:;;: b, has a maximum value (see Section 3.5). This result
is intuitively "obvious" to most students, but present-day standards of mathe-
matical rigor demand strictly logical proofs even for "obvious" results. There are
several good reasons for insisting on absolute rigor in mathematics, especially
in the foundations. One reason is that seemingly obvious results sometimes
turn out to be wrong or, more frequently, imprecise. When more complex
situations are studied (for example, calculus in several dimensions), things may
no longer be so obvious. Thus we may be forced to analyze the simplest case in
great detail before attacking the more complex case. An innocent error at the
initial level may become serious when compounded in a complex problem.
The need for absolute, logical precision in mathematics has been recognized
since Greek times as a characteristic aspect of the subject. Logical standards
have in fact become steadily more rigorous throughout the history of mathe-
matics. As a consequence, mathematics is toduy the least controversial of all
subjects studied by man. Working scientists virtually never find it necessary to
question the (.!orrectness of the mathematics which they use. (Scientists do
91
92 Chapter Three Properties of Continuous Functions
sometimes complain that mathematicians are too slow in developing the

mathematics they require. Unfortunately the' required mathematics often
happens to be either very difficult or very boring.)
The theorem stated earlier-that a continuous function on a ::;; x < b has
a maximum value-is an example of what mathematicians call an existence
theorem. It tells us that a maximum value exists, but not how to find it. The
importance of existence theorems can be demonstrated by the following simple
example.
Example
Suppose we wish to prove: "I is the largest positi:ve integer."

Let x denote the largest positive integer. Then x ~ 1, so that x 2 ~ x. But
x is also a positive integer. Therefore x 2 = x. Dividing by x, we obtain x = 1.
2
What is the error in this "proof"? The moral of this example is that if we
refer to nonexistent objects as if they existed, we may be led into foolish errors.
Mathematicians seem to have learned this moral; politicians probably never
will.
Exercises
1 +X~
1. Let x 1 = 2 and Xn+l = -- - , for n >- 1. Let limn-co Xn = a. Then
2
2a = 1 + a2 , so that a = 1. What is wrong with this?
2. Criticize the following "ontological" proof of the existence of God:

(a) By definition, God is the greatest conceivable being;
(b) If God did not exist, then it would be possible to conceive of a greater
being (that is, an existent being). This is a contradiction, so God must
exist.
3.2 Supremum and Infimum
In this section we discuss the important concepts of the supremum (sup)

and infimum (inf) of a set A of real numbers. Under special circumstances the
supremum is called the maximum (max) of A and the infimum is called the
minimum (min) of A.
Definition 1 Let A be a given subset of &. A number A....i called an~

bound o(A (respectively, a lower bound of A) i(x >a (or every a E A (respec-
tively, x :::;; a for every a E A).
/}efinition 2 Let A be a subset of &. A number M is called the SUJ!remum (or

least upper bound) of A if
..ill. M is an upper bound of A; and

..ill} if x is any upper bound of A, then we have M :::;; x.
-
Similarly, m is the infimum (or greatest lower bound) .oJ..A.JJ.
.ill.. m is a lower bound of A; and
..@2 ify is any lower bound of A, then we have m ~ y.
The standard notation is:
M = supA, m = inf A;
the notation M = lub A and m = glb A is also common. If A has no upper

bound, we write
sup A= +oo.
Similarly if A has no lower bound, we write inf A = - oo.
Example I
Let A = r: I IX > o}.t Find inf A and sup A.

. x +I
Solution Smce - - = I
X
+ -XI ~ I, we see that 1
.
IS a lower bound for
A. Clearly no number y > I can be a lower bound for A (why not?). Hence,
inf A = I . Since A contains arbitrarily large numbers, sup A = + oo.
t The notation { y J } stands for the set of ally such that ....
Example 2
Let B = {x+11
-x- x IS . a positive
.. mteger
. }. The reader can easily verify that
sup B = 2 and inf B = 1.
Notice in this example that sup B is a member of the set B, but inf B is not.
D(/inition 3 I{ sup A is a member of the set A. we say that sup A is the

maximum of A:
sup A- max A:
in this case we also say that sup A is assumed in A. Similarly we write
inf A- min A
in case inf A belongs to A.
In most applications in this text, the set A will be the set of values of some
I
function: A= {f(x) x E B}, where B is some subset of the domain off In this
case we use the notation
. sup f(x) = sup {f(x) I x E B}.

o:eB
Example 3
1
Ifj(x) = - -2 , find
1+x
sup f(x), inf f(x), maxf(x), and minf(x).

O:EiiJt O:EiiJt
Solution It is easy to see that
1 1
sup--=
2
max--= 1
o:eiilt1+x2 o:eiilt1+x
and
1
inf-- =0
o:eiilt 1 + x 2 '
but this is not assumed, so that minxe9l __ I_ does not exist. See Figure 3.I.
I + x2
We remark that in general M- max.,.,n [(x) if and only if (i)f(x) < M
for all x E Band (ii) there exists a point x 0 E B such that f(x 0 ) = M.
Example 3 shows that a function may be bounded and still not have both
a maximum and a minimum value. It seems reasonable however that for any
\inf f(x)
Figure 3.1 f(x) = I/(I +x 2

).
bounded function f(x) both supxeB f(x) and inf.,EB f(x) will exist. This is
indeed the case, and the completeness property of f!ll is required in the proof.
Theorem Let A be a nonempty subset of f!ll and suppose A has an upper

bound Then sup A exists (and isftnite).
This theorem can be reworded as follows: every nonempty subset A of f!ll

which has an upper bound, has a least upper bound. In many texts this state-
ment is assumed as an axiom about the real-number system. It can be shown
(see Exercise I2) that this statement is logically equivalent to the completeness
property of f!ll. Thus it is simply a matter of taste which statement is used as an
axiom. In fact, as we mentioned in Chapter I, both statements can be proved
as theorems in a more painstaking development of the real number system.
Proof of Theorem We will construct a nondecreasing sequence {xn} of points

of A, and a nonincreasing sequence {yn} of upper bounds of A, in such a way
that
limxn =limyn =A. (3.1)
n-+co n-too
Then we will show that A = sup A.

To begin;let x 1 be any point of A and let y 1 be an upper bound of A; both
x1 and y 1 exist by assumption. Obviously x1 :::;;: y 1. Let
z1 = Hx1 + Y1),
so that z1 is halfway between x1 and y 1. Determine x 2 and y 2 as follows. If z1 is an
upper bound of A, let x 2 = x 1 and Y2 = Z 1 If z 1 is not an upper bound of A,
let x 2 be some point of A with x 2 > z1 , and let y 2 = y 1 In either case we have:
(a) x 2 E A and y 2 is an upper bound of A;
(b) x1:::;;: x2 :S::y2 :S::y1;
(c) Y2 - X2 :S:: HY1 - x1).
Having determined x 2 and y 2 , we can repeat the process to obtain points x 3 , y 3
Repeating the process indefinitely (by induction), we obtain two sequences:
a bounded, nondecreasing sequence {xn} of points of A and a bounded, non-
increasing sequence {yn} of upper bounds of A. By the completeness property,
both of these sequences must converge. Since clearly
as n ~ oo, the sequences must have the same limit A. This proves Equation (3.1).
We show finally that A = sup A. First set x EA. Then Yn ;;::::: x for all n
because each y n is an upper bound for A. Therefore A = limy n ;;::::: x. This shows
that A is an upper bound for A. Next, if p, < A, then xn > p, for large n since
lim xn = A. Since xn E A, this shows that pis not an upper bound for A. There-
fore A is the least upper bound of A: A = sup A. I
Corollarr. Let f(x) be a function defined for all x E B. Then sup.,en f(x)
+
always makes sense,either as a finite number, or as oo. Similarly, infxeB f(x)
.
is either finite or - oo .
Definition 4 If supwd? f(x) < oo we say that f(x) is hounded above on B,

whereas ifinfxen f(x) > -oo we say that ((x) is hounded below on B . lff(x) is
bounded both above and below on B. we say simply that ((x) is hounded on B .
Exercises
I. Find sup A and inf A for each of the following sets A. Determine whether
max A and min A exist.
(a) A = {x2 1 X E R}.
(b) A is the finite set {1, ~~, 2

7T , 10} (Note: 3.1415 < 7T < 3.1416).
(c) A is the open interval (0, 1) = {xI 0 < x < 1}.

(d) A is the half-open interval (0, 1] = {xI 0 < x :::;;: 1}.
(e) A is the set of all real numbers in (0, 1) whose decimal expansion con-
tains no 9's.
2. (a) Let 0 denote the empty set, that is, the set that has no elements. What
numbers are upper bounds for 0? Does sup 0 exist?
(b) Show that every subset of fJe has a supremum and an infimum, provided
that the "values" oo are permitted.
3. Find the following sups and infs. State which are maxs and which mins.
(Use calculus if necessary; a sketch of the function may also help.)
1 1
(a) sup--, (b) inf - - ,
z>O 1 +X x>O 1 +X
(c) inf
z>O
(x + .!) ,X
(d) sup x and inf x ,

zeR J 1 + x2 zeR J 1 + x2
sin x
(e) sup-, (f) inf sin x ,,I
"'*O X "'*O X
(g) inf sin x (You may not be able to solve this exactly!).
z*O X
4. Explain why sup A is unique if it exists.
5. What if sup A = inf A?
6. Let A = I
{nll" n is a positive integer}. Find sup A and inf A.
7. Letf(x,y) = x + 2y. Find

and inf [ sup f(x,
1~11~3 o~z~l
y)J.
8. Prove that M = sup A if and only if

(a) M is an upper bound of A, and
(b) there exists a sequence {x..} in A with limn-.oo x .. = M.
9. Prove that
sup [f(x) + g(x)] :::;; sup f(x) + sup g(x). (3.2)
o:eA xeA xeA
(Hint: Let M = supxeAf(x) and N = SUPxeA g(x). Then f(x) :::;;; M and
g(x) :::;;; N for every x EA. Complete this argument.)
10. Give examples of functions f(x) and g(x) such that
sup [f(x)
o;e[O,l)
+ g(x)] < sup f(x)
o;e[O,l)
+ sup g(x);
o:e[O,l)
also give examples in which this is an equality.
11. State and prove the analogous inequality to (3.2) for the case of the
infimum.
*12. Assume that f!ll satisfies conditions 1-10 of Section 1.3, as well as
11 '. Every nonempty subset A of f!ll that has an upper bound has a least
upper bound.
Prove that f!ll has the completeness property. (Hint: If {x..} is a non-
decreasing bounded sequence, then A.= sup.. ;:: 1 x .. exists. Show that
A = limn-.oo x,..)
3.3 The Bolzano- Weierstrass Property
The completeness property tells us that a bounded monotonic sequeDce

must converge. In this section we show that epery bounded sequence must have
some convergent subsequence.
JJe{inition Let S be a given subset of f!ll. A point Xo E f!ll is called an accumu-

lation point o{S provided that every e-interval o(x0 (e > 0) contains some point
oJS not equal to x 0 itself.
It can easily be shown that Xo is an accumulation point of S if and only if

there is some sequence {xn} in S such that lim,,....w x,. x0 aDd Xn =!= x0 for
3.3 The Bo/zano-Weierstrass Property 99
~ (see Exercise 4). You should verify the following simple examples:
~ A finite set has no accumulation points.
JQl_ If S = (0, 1), then the set of all accumulation points of Sis the closed
interval [0, 1].
() If S = !2 (the rationals), then eve:.y real number is an accumulation
point of S.
Theorem (Bolzano-Weierstrass) A bounded, infinite set of real numbers has

at least one accumulation point.
Proof Let S be a given bounded, infinite set; then there is some interval
[at> b1 ] containing S. Let
Since Sis an infinite set, at least one of the intervals [a1 , c1 ] or [c1 , b1 ] contains
infinitely many points of S. Let [a2 , b2 ] denote such an interval. Note that
The same process may be repeated inductively to obtain a sequence of intervals

[an, bn], each containing infinitely many points of S, and such that
and
(3.4)
According to the completeness property, a= limn~co an and b = limn~co bn

exist; (3.4) obviously implies that a =b. We assert that a is an accumulation
point of S. To see this, suppose e > 0 is given. Then for sufficiently large n we
have
Since [an, bn] contains infinitely many points of S according to the construction,
the interval (a - e, a+ e) must surely contain some point of S not equal to a.
This completes the proof. I
Notice that the proof of the Bolzano-Weier<;;trass theorem once again used
the completeness property of !Jl. Conversely, it is possible to prove the com-
plet~ness property of fJl if one assumes the validity of the Bolzano-Weierstrass
theorem; see Exercise 5.
100 Clwpter Three Properties of Continuous Functions
Corollary Let {xn} be a sequence of real numbers. If {x.,} is bounded. then

{x.,} has a convergent subsequence.
' I
Proof Let S denote the set of all values Xno n = I, 2, 3, .... There are two
cases to consider: either S is a finite set or S is an infinite set.
In case S is a finite set, there must be infinitely many values of n, say
n1 , n2 , n3 , , such that xnk = x where x is one of the members of S. Hence,
{xn} has a convergent subsequence.
If Sis an infinite set, we apply the Bolzano-Weierstrass theorem to obtain
an accumulation point a of S. It follows (cf. Exercise 4) that some subsequence of
{xn} must converge to a. I
Exercises
1. Find the set of all accumulation points for each of the following sets
I
(a) S = {x E &l' 0 < lxl < 1}.
(b) Sis the set of all integers.
(c) Sis the set of all numbers m/2n, where m, n are integers.
(d) Sis the set of all rational numbers p/q in which lpl < 10.
2. Let S be a bounded infinite set in &l'. Show that either sup S = max S or
sup S is an accumulation point of S. Give examples for both possibilities.
3. If {xn} is a given sequence, let S denote the set of values of xn, n = I, 2,

3, ... , and let S' denote the set of accumulation points of S. Find examples
of sequences {xn} for which
(a) S' is empty,
(b) S' consists of one point,
(c) S' consists of two points,
(d) S' is countable,
(e) S' is uncou~table.
4. Prove that x is an accumulation point of a set S if and only if there is a

sequence {xn} of points of S such that Xn =f- x for all n, and limn~oo Xn = x.
5. Assuming that the real-number system &l' has Properties l-10 of Section
3.1, and that the Bolzano-Weierstrass theorem is valid, prove that the
completeness property holds in &l'.
3.4 Cauchy Sequences 101
3.4 Cauchy Sequences
We often wish to know whether a given sequence {xn} converges. Sometimes

we can calculate Iimn~oo Xn by inspection or by using simple theory. If {xn}
is increasing or decreasing we can appeal to the completeness property. But
many other sequences can arise. For many years mathematicians tried to find a
necessary and sufficient condition for the convergence of a given sequence
{xn}. Some people asserted that Iimn~co (xn+l - Xn) = 0 was such a condition,
but we know this is false (cf. Exercise 1). The problem was solved by A. Cauchy.
Definition A sequence {xn} of real numbers is called a Cauchy sequence if it

satisfies the following condition:
fnr any given e > 0 there exists an integer N (which may depend on e)
such that
lxn - xml < e for all n, m :::0: N. (3.5)
Notice that although this definition is reminiscent of the definition of the

limit of a sequence, it refers only to the terms of the sequence {xn}, and not to
any anticipated limit L.
Theorem (Cauchy) A necessary and sufficient condition {or convergence of a

sequence {xn} is that it be a Cquchy sequence.
Proof The proof of necessity is very easy and will be given first. Suppose that
Iimn~coxn = L exists. Let e > 0 be given, and choose an integer N such that
lxn - Ll < e/2 if n :::0: N. Then, for both n, m :::0: N, we have
which shows that {xn} is a Cauchy sequence.

The converse proof is slightly more difficult. First we will show that a
Cauchy sequence is bounded. For this purpose let e = I; then there exists an
integer N such that lxn - xml < I for n, m :::0: N. In particular lxn- xNI < I
for n :::0: N, which shows that the sequence {xn} is bounded for n :::0: N, and
hence for all n.
Now by the corollary to the Bolzano-Weierstrass theorem, the bounded
sequence {xn} must have a convergent subsequence {xnJ Let Iimk~co xn. = L.
For any integer n we have
lxn- Ll :S::: lxn- Xnkl + lxnk- Ll.
Given e > 0 we can find an integer N such that lx, - xml < ef2 for all n, m 2
N, and also lx,k - Ll < e/2 for k 2 N. Therefore, if n 2 N, we can choose a
term x,k such that
This shows that {x,} converges to L. I
The condition (3.5) is often called the Cauchy criterion for conyergen~e;
it is sometimes written as
lim (x, - Xm) = 0. (3.6)
n,m-tco
For infinite series Cauchy's theorem immediately implies a corollary.
.
Corol/arv An infinite series !f a, converges if and only if
m
lim ---! ak = 0.
n,m-+co k=n
Exercises
1. Show that if {x,} converges, then lim,__,. 00 (x,+I - x,) = 0, but not con-
versely.
1
2. Suppose that lx,+ 1 - x,l :::;: , for all n. Prove that {x,} converges.
2
3. Let f(x) be defined for x > 0. Show that lim.,__,. 00 f(x) exists (finitely) if
and only if for every given e > 0 there exists a number M such that
lf(x) - f(y)i < e whenever x 2 Mandy 2 M.
(Hints: Necessity is easy. For sufficiency, apply Cauchy's theorem to

the sequence {f(n)} to prove that lim,__,. 00 f(n) = L exists. You still have
to show that lim.,...... oo f(x) = L.)
4. An infinite series !fan is said to converge absolutely if the series !f Ia,!

converges. Use the corollary in this section to give a simple proof that
absolute convergence implies ordinary convergence.
5. Give an example of an infinite series which converges, but does not con-
verge absolutely.
3.5 Properties of Continuous Functions
In this section we prove several basic theorems about continuous functions.

Although these theorems are often referred to in calculus texts, their proofs are
usually considered "beyond the scope" of such texts.
In each of the theorems, it will be supposed thatfis a function that is defined
and continuous on a closed, finite interval [a, b].t Remember that~
this means that f is continuous at each point x 9 in [a, b]. Strictly speaking we
have not yet defined continuity at an end point. This omission is easily rectified:
when f is defined on a closed interval [a, /;] we say that f is continuous at
a provided
lim f(x) = f(a),
a:-+a
and similarly f is continuous at b provided
lim f(x) = f(b).

a:-+b-
Example
The functionj(x) = -./1 - x 2 is continuous on the closed interval [ -1, 1].
,.
Theorem I (Intermediate value theorem) Let f be defined and continuous on
the closed finite interval [a, b] . Suppose that
f(a) < 0 <f(b).
Then there exists a point c jn the open interval f.fhl22 such that ((c) = 0 (see
Figure 3.2).
t Henceforth we adopt the modern notation for functions. A single symbol (f,g, 'Y, etc.)
will denote a function; the notation f(x) will be used to denote the value of the function fat the
point ;c. (Nevertheless we will also continue to use expressions such as "the function f(x) = x"
rather than the pedantic form "the functionfwhose value at xis x 2 ."}
104 Chllpter Three Properties of Continuous Functions
Figure 3.2 The intermediate value theorem.
Proof The idea of the proof is to define c as the "largest" value of x for which
f(x) ::::;; O,'namely, --
c =sup {x E [a, b] lf(x) ::::;; 0}.
That such a number c exists follows from Section 3.2: the set {x E [a, b] If(x) ::::;;
0} is nonempty (because f(a) < 0) and bounded above by b. We 'wish to show
thatf(c) = 0. Suppose thatf(c) < 0. By continuity ofjat c there exists {J > 0
such thatf(c +b)< 0. But this contradicts the definition of c.
Similarly if f(c) > 0, then there exists {J > 0 such that f(x) > 0 for all
x > c - {J (note that f(x) > 0 for all x > c by the choice of c). This again
contradicts the definition of c. Thereforef(c) = 0. Finally, sincef(a) < 0 and
f(b) > 0, we cannot have c = a orb, so that c E (a, b). I
In general, ifjis continuous on [a, b], thenf(x) must assume every value
betweenf(a) andf(b) at some point x between a and b. This is easily proved
from Theorem 1 (see Exercise 2).
CorollarY. Any polynomial o{odd degree has at least one real root.
Proof Let p(x) = aoxk + + ak be such a polynomial, where k is odd and,

without loss of generality, a0 > 0. We know that p(x) is continuous for all x
and also that
lim p(x) = + oo and lim p(x) = - oo.
x~+oo x~-oo
Hence, we can choose some interval [a, b] such that p(a) < 0 < p(b), and there-
fore p(x) = 0 for some point x. I
We now establish another fundamental property of continuous functions.

First we state a useful lemma.
Lemma Suppose that f is continuous at x 0 Then if {x,} js any sequence

converging to x 0 , it follows that the sequence { ((xn)} converges to ((xo).
The proof is a simple e-b argument, which we leave to you to carry out.
The converse of this lemma is also valid: if I (x,)---+ I (x 0 ) for every sequence
{x,} which converges to x 0 , then lis continuous at x0 ; see Exercise 8.
Theorem 2 Let fbe defined and continuous on the closed finite interval [a, b]
Ihenfis bounded on [a, b], that is~ . -
sup lf(x)l < +oo.

[a, b)
Proof We prove this by contradiction. Suppose I is not bounded on [a, b] ;

for example, suppose I is not bounded from above. Then for any positive
integer n there must exist a point x, in [a, b] such thatl(x,) ~ n. According to
the Bolzano-Weierstrass theorem, the sequence {xn} must have a convergent
subsequence {x,k}; let x 0 = limk-oo x,k. Then x 0 E [a, b]. Since I is continuous
at x 0 , we must have, by the lemma,
limf(xnk) = f(xo).
k-+oo
But since l(x,t) ~ nk, we have instead
Iimf(x,k)
k-+oo
= + oo.
This contradiction completes the proof. I
Theorem 3 Letfbe defined and continuous on the closed finite interval.la..lz1

Then max[a.bJ f(x) and minra b) ((x) both exist. (In other words, l(x) assumes
its maximum and minimum values on [a, b].)
Proof Let M = sup[a,bJ l(x); by Theorem 2weknowthat Mis a finite number.

It follows from the definition of the supremum that there is a sequence {x,}
in [a, b] such that l(x,)---+ Mas n---+ oo (see Exercise 8, Section 3.2). By the
Bolzano-Weierstrass theorem, this sequence has a convergent subsequence
f(x 0 ) =lim f(xnk) = M.

k--+oo
Therefore M = max[a.b] f(x) . Similarly
m = inf f(x) = minf(x). I

[a,b] [a,b]
Theorem 4 Let f be defined and continuous on the closed interval [a, b], and
suppose ((x) ::F 0 (or all x E [g. bl. Then, either f(x) > 0 for all x E [a. b] Q!.
[(x) < 0 (or all x E [a, b].
In case f(x) > 0 for all x in [a, b ], there exists a positive number t5 such that
f(x) ;;;:=:: t5 > 0 for all x in [a, b].
This theorem is an immediate consequence of Theorems I and 3. We leave

the details to the reader. Sometimes the theorem is stated briefly as follows:
"a continuous function which does not vanish on a closed, finite interval,
is bounded away from zero on the interval."
Exercises
I. Letf(x) = x - [x], where [x] denotes the greatest integer in x . Show that
althoughfis bounded, max[ 0 , 21 f(x) does not exist.
2. Prove the generalized intermediate value theorem: iff is continuous on

[a, b], thenf(x) assumes every value betweenf(a) andf(b) at some point
of [a, b]. (Hint: If ~ lies between f(a) and f(b), apply Theorem I to the
functionf(x) - n
3. Show that iff is a continuous function defined on [0, I] and ifO ~f(x) ~ I
for all x, the equationf(x) = x has a solution in [0, I].
4. Let p(x) = x 4 + a 1x 3 + a 2x 2 + a3x +a 4 Show that if a4 < 0, then p(x)

has at least two distinct zeros.
5. Show by examples that a continuous function defined on an open, finite

interval need not be bounded, and even if bounded, need not have a maxi-
mum value on the given interval.
6. Letfbe continuous on [a, b], a< b. Supposefis also one-to-one, that is,
x 1 =f=. x 2 implies f(x 1 ) =f=. f(x 2 ) . Prove that f must be a monotone function.
*7. Letfbe a strictly increasing, continuous function on [a, b]. Then there is a
uniquely defined inverse function g defined on the interval [f(a), f(b)]
(that is, g(f(x)) = x for all x in [a, b]). Prove thatg is continuous.
8. (a) Prove the lemma of this section in detail.

(b) In terms of e-b, what does the statement lim.,_.a f(x) =f=. L mean?
(cf. Appendix I, Section 2.)
(c) Prove the converse described following the lemma.
3.6 Uniform Continuity
We now consider the concept of continuity a little more deeply. Let f

be a function defined on an interval I; we allow the possibility of infinite intervals
I. We have definedfto be continuous on I iff is continuous at each point of I.
From the definition of continuity at a point x, we therefore have: f is continuous
on I if,for every x E I and for every e > 0, there exists a number b > 0 such that
if(x') - f(x)i < e if ix' -xi < b. (3.7)
Example 1
Letf(x) = x 2 on I= ( -oo, oo). Calculate b so that (3.7) is valid.
Solution First let b :::;; 1. Then ix' - xi < b will imply that
ix' + xi :::;; ix' - xi + i2xi :::;; 1 + 2 ixl.

Therefore
lx' 2 - x2 i = ix'- xlix' +xi
:::;; ix'- xi (1 + 2ixl) (iflx'- xi:::;; 1)
<e (iflx'-xi< e)
1 + 2l xi
Hence, we may choose
(3.8)
108 Chapter Three Properties of Contillllous Functions
Notice that this value of <5 depends on x; as ixi increases, the value of <5 given
by (3.8) decreases. Consideration of the graph y = x 2 (see Figure 3.3) makes it
clear that (for a given e) the value of <5 required in (3. 7) necessarily decreases as
ixi increases. Moreover no fixed value of <5 > 0 will work for all x in ( - oo, oo).
Figure 3.3 f(x) = x 2
Example 2
Iff(x) = x 2 is considered only on the bounded interval 0 :::;:; x :::;:; 3, then a value
~independent of x can be found so that (3.7) holds.
To see this, we simply note that if 0 :::;:; x :::;:; 3, then I + 2 ixi :::;:; 7. Hence,
by the same calculations as in Example I,
if(x') - f(x)i = ix' 2 - X

2
i :::;:; ix' - x i (1 + 2ixi) (if ix' - xi :::;:; 1)
:::;:; 7ix' - xi
< e (if also ix' - xi<~)
Hence we may choose <5 = min (I, e/7), and this is independent of x.
A function f defined on an interval I will be called uniformly continuous

on I if the value o in (3.7) exists and can be chosen independently of x E I.
More precisely, we have:
Definition Let f be defined on an interval I. Then f is said to be uniformly

continuous on I if the following condition holds:
for any given e > 0, there is a corresponding fJ > 0 such that
(3.9)
lf(xt) - f(x 2)1 < e for all points xi> X 2 E I with lx1 - x 2 1 < fJ.
We emphasize that the number fJ may depend on e, but not on x 1 or x 2

(For simplicity of use, this definition has been made symmetric in the values of
x appearing in (3.7), which we now denote by x 1 , x 2 instead of x, x'.)
We can now prove the following important result.
Theorem Let f be defined and continuous on a closed, finite interval I.

Then f is uniformly continuous on I.
Proof The proof will be by contradiction; suppose f is not uniformly con-

tinuous on I. Negating condition (3.9) (cf. Appendix I), we therefore have:
there exists a number e > 0 such that for every fJ > 0 there are points
x, y E I (depending on fJ !) such that
lx - yl < fJ and lf(x) - f(y)l ;;-:.; e.
We will apply this statement to the values (J = 1, 1/2, 1/3, ... , 1/n, .... For
fJ = 1 we obtain points x 1 , y 1 E I such that
In general, for every integer n 2 1, we obtain points Xn, Yn E I such that
(3.10)
Since I is a closed, finite interval, the Bolzano-Weierstrass theorem

implies that the sequence {x11 } has a convergent subsequence {xnJ Let x 0 =
limk-co Xnk. Sincelxnk- Ynkl < 1/nk-+Oask-+ oo,itfollowsthatlimk..... coYnk=
x 0 also.
By hypothesisf(x) is continuous at x 0 Therefore
lim f(xn)
k
= f(x 0 ) =lim f(Y 11k ) .
k-+co k-+ co
110 Clulpter Three Properties of Continuous Functions
Consequently we can find an integer nk such that
e e
<-+-=e.
2 2
This contradicts (3;10), and the proof is complete. I
Exercises
1. The following functions are uniformly continuous on the indicated intervals.

Calculate a value of t5 independent of x1 , x 2 such that (3.9) is valid.
(a) f(x) = x : ; I = [0, 3],

1
{b) f(x) = 2x2 - x; I= [0, 10],
(c) f(x) = 3x-1 ; I= [1, oo).
2. Show that a linear functionf(x) =ax+ b is uniformly continuous on any

interval I.
3. The functionf(x) = Ifx is continuous on the open interval (0, I). Show that
it is not uniformly continuous there.
4. Isf(x) = lxl uniformly continuous on (-oo, oo)?
5. Letf(x) =sin 1r/x be defined on I= (0, I). It should be obvious thatfis

not uniformly continuous on I. Where does the proof of the Theorem of this
section break down?
6. Prove that iff is uniformly continuous on the open, bounded interval

I= (a, b), thenfis bounded on I. Does it follow that max1 f(x) exists?
7. Let f(x) = 1/(l + x 2). By determining a number t5 independent of x,

show thatfis uniformly continuous on ( -oo, oo).
8. Let f(x) be uniformly continuous on (a, b) and let {xn} be a sequence of

points in (a, b) converging to a. Prove that limn-oo f(xn) exists. (Hint:
Use Cauchy's theorem, Section 3.4.)
4 Some Theorems of Calculus
On the basis of the theorems of Chapter 3 we can now give rigorous proofs
of certain basic theorems of elementary calculus.
4.1 The Mean Value Theorem
The proof of the mean value theorem (for derivatives) begins with the
following special case.
Lemma (Rolle's theorem) Let f be defined on [a, b,] where a < b. Assume that
(i) f is continuous on [a, b ],

(ii) f is differentiable on (a, b),
(iii) f(a) = f(b) = 0.
Thenf'(e) - Ojor some pointE E (a, b).
Proof If f(x) =0 on [a, b] the theorem is trivial. Otherwise we can assume

thatf(x) > 0 for some point x in (a, b). Sincefis continuous on [a, b], it must
e
therefore have a positive maximum at some point in [a, b] (see Figure 4.1),
according to the results of Chapter 3 (namely, Section 3.5, Theorem 3). By
e e
hypothesis (iii), =/=a or b, so that E (a, b). Since f has a maximum at e,
111
111 Chapter Four Some Theorems of Calculus
we havef(~ +h)-/(~) ::::;; 0 for all h. Consequently,
f(~ +h)- f(~) {::::;; 0 if h > 0,

h ::?: 0 if h < 0.
Since
!'(~) = lim f(~ + h) - !(~)
h-+0 h
exists by hypothesis, we conclude that both/'(~) ~ 0 andf'(~) ::::;; 0. Therefore,

!'<E)= o. I
y
Figure 4.1 Rolle's theorem.
To derive the mean value theorem from Rolle's theorem, we consider a

function f satisfying hypotheses (i) and (ii) of Rolle's theorem, but having
arbitrary values at x = a and x = b. Let l(x) be the linear function passing
through the points (a,f(a)) and (b,f(b)); see Figt!re 4.2. Then
l(x) =f(a) +f(b)- f(a) (x- a). (4.1)

I
b-a
a ~
Figure 4.2 The mean value theorem.

4.1 The Mean Value Theorem 113
Now let g(x) = j(x) - l(x); it is easily seen that g(x) satisfies all three hypoth-
e
eses of Rolle's theorem. Consequently there is some point E (a, b) such that
g'(e) = f'<e) - l'(e) = J'(e) - f(b) - f(a) = o.

b-a
We have therefore proved the following theorem.
,Theorem (The mean value theorem) Let f be defined on [a, bJ, where a <b
Assume that
I(i) f is continuous on [a, b],
t ii) fis differentiable on (a, b).
Then there exists some point eE (a, b) such that
f(b)- f(a)
f'(e) (4.2)
b-a
The mean value theorem has the foliowing two corollaries, which calculus
students usually consider to be "obvious."t
Corollarv I Suppose that f is defined on (a, b) and that f'(x) = 0 (that is,
f'(x) = 0 for every x in (a, b)). Then [(x) is con.~tant on (a, b).
Corollarv 2 Suppose that f is defined on (a, b) and thatJ.1& > 0 (or all x in
(a, b). Then (is strictly increasing on (a, b).
The proofs are fairly "obvious" consequences of the mean value theorem,
and are left for the exercises.
We will use the mean value theorem and its corollaries in studying integra-
tion in Section 4.3. Before leaving this theorem, let us comment briefly on its
hypotheses. Notice that f is required to be continuous on the closed interval
[a, b] and differentiable on the opgn i nter~L(g~}. If either of these hypotheses
is violated. the conclusion fails (see Exercises 6 and 7).
t Professional mathematicians use the word "obvious" to indicate that it is obvious

how to give a complete proof. To use "obvious" to mean "I'm sure it's true, but I can't prove
it," is not a commendable practice.
Exercises
I. Letf(x) = ..;--;;, 0 ::;:: X ::;:: a. Find all points eE (0, a) for which the mean
value formula,
f'(e) = f(a) - f(O) = f(a) ,
a -0 a
is valid. Does f satisfy all hypotheses of the mean value theorem?
2. Prove Corollary I. (Hint: If a< x 0 < x 1 < b, prove that j(x0) = f(x 1).
This proves the corollary; why?)
3. Prove Corollary 2.
4. Prove: Ifj'(x) ::2:0 on (a, b), thenf(x) is nondecreasing on (a, b).
5. State analogues to Corollary 2 and to Exercise 4, for the cases f' (x) < 0,
f'(x) ::;:: 0, respectively. Give one-line proofs of these analogues, without
using the word "similarly."
6. Let j(x) = I - lxl be defined on [-I, I]. Show that f satisfies neither the
hypotheses nor the conclusion of Rolle's theorem.
7. Give an example of a functionfwhich is differentiable on (0, I), but not )

continuous on [0, I], for which the conclusion of the mean value theorem '-
fails.
8. Give an example of a functionfsatisfying the hypotheses of Rolle's theorem

- on [0, I] and having infinitely many points g satisfying the conclusion of
Rolle's theorem.
9. Use Rolle's theorem to show that the polynomial equation r +X- 9 = 0

has exactly one real solution.
10. Show that if a > 0 and b is an arbitrary real number, then the equation
x 3 + ax + b = 0 has exactly one real solution.
II. Show that !sin x - sin yl :S:: lx - y!.
I2. Prove that iff' (x) = g' (x) on (a, b), then j(x) = g(x) + con st. on (a, b).
(Give a very short proof!) j
13. Show thatj"(x) = 0 impliesf(x) =ax+ b. I

I
cn.J
X 14. Letfbe defined on ( -oo, oo) and suppose
sup lf'(x)l < oo.

-oo<x<oo
Show thatfis uniformly continuous on ( -oo, oo).
4.2 The Riemann Integral
The definition of integral given in this section is due to G. Darboux (1842-

1917); it is equivalent to a definition given earlier by B. Riemann (1826-1866).
The motivation for the definition is, ofcourse, that the definite integral
J:J(x) dx
should represent the "area under the curve" y = f(x) for a ::;; x ::;; b.
Definition 1 Let [a, b] be a given (finite) interval. A partition 7T of [a, b] is a

finite set of points {x0 , x 1 , , Xn} with
a= x 0 < x 1 < < Xn =b.

we use the notation
~xk = xk - X~c- 1 (k = 1, 2, ... , n).
1
A second partition 7T of [a, b] is said to be finer than 7T if 7T 1
contains all the points
of 7T. In this case, we write 7T > 7T.
1
Definition 2 Let f be a bounded function diffined on [a, b] and let 7T be a

partition of[a, b]. Then we define the upper Darboux sum off relative to 7T ~:
(4.3)
Similarly, the lower Darboux sum off relative to 7T is defined to be
(4.4)
116 Cluzpter Four Some Theorems of Calculus
The geometric meaning of these expressions is illustrated in Figure 4.3,

in which the shaded areas represent the upper and lower Darboux sums off
relative to the given partition. It should also be clear from the figure that if
A = area under the curve y = f(x), for a :::;; x :::;; b,

then
(4.5)
This statement cannot be proved as a theorem, for the simple reason that
we do not as yet have a definition for the word"area." In fact, we are going to use
the inequalities (4.5) to define area.
~~~ =upper sum II= lower sum
Figure 4.3
Definition 3 Let f be a bounded function on [a, b]. If there exists a unique

number A such that (4.5) holds for every partition 7T of [a, b], we write
A =t a
f(x) dx (4.6)
and call A the area under the curve y = f(x) for a :::;; x :::;; h. In this case we say
that f is integrable over [a, b] and A = S! f(x) dx is called the (definite) integral of
f over [a, b].
We will see shortly that numbers A satisfying (4.5) always exist; it is im-
portant to realize that A must be unique in order for us to define J! f(x) dx = A.
Example 1
Letf(x) = x, 0 :c:;; x :c:;; I. It is "obvious" that A= 1/2. Let us just check that
this value A = 1/2 is the only possible value of A for which (4.5) can hold for
every partition 7T (Figure 4.4).
n
Suppose 7T = I 2
{0,-n , -n , ... , -n = I }.. Then ~xk = 1/n. Also
r-
1 I
-- _ _J
I I I
I I I
.-- ----l I
I I I I
I I I I
.-- --i I I
I I I I I
I I I I I
- -~ I I I
I I I I I
Figure 4.4
Substituting these values in (4.3), we find that
Similarly,
L,(f) = i
k=l
k- 1
n2
= n-
2n
1.
Hence, by (4.5), A must satisfy
n-1:c:;;A:c:;;n+1.
2n 2n
Since n is arbitrary, it clearly follows that A = 1/2.

Next we give an example of a function fwhich is so "bizarre" that the con-

cept of "area under f" loses meaning.
Example 2
Let
1 if x is rational,
f(x) = { . . . .
0 1f x 1s urahonal,
and consider the interval [0, 1]. If 0 ~ X~c- 1 < xk ~ 1, then
sup f(x) = 1 and inf f(x) = 0.
Therefore, for any partition rr we have
n
U,,(f) = !1 (xk- X~c-J = 1,
k=l
n
L,(f) =!
k=l
0 (xk - xk-t) = 0.
Thus the inequalities (4.5) are satisfied for any number A with 0 ~A ~ 1.
This means that f is not integrable over [0, 1].
Recall that the function of Example 2 is everywhere discontinuous (Section

2.4, Example 7). It is known that no such function can be integrated by the
Riemann method.t
We come now to the elementary theory of the Riemann integral. Hence-
forth, let f denote a given bounded function on [a, b].
Lemma Le~ rr1 , 17 2 be partitions of [a, b]. Then

(i) L, 1 ( / ) ~ u".(f);
(ii) if 1r1 < rr2 , then
t Cf. R. C. James, Advanced Calculus, Wadsworth (1966), p. 150.

4.2 The Rie11111nn Integral 119
Proof We prove (ii) first. Suppose to begin with that 1r 2 contains just one more
point x' than 1r1 , and let xk-l < x' < xk, where X~c- 1 , xk are points of 1r1 Then
we have (see Figure 4.5)
sup f(x)}(x' - xk_1)

{["'k-1"''1 + { sup f(x)}(xk - x')
["'' ,.,.J
~{ sup f(x)}(x' - xk_1 + xk - x')

["'k-l"'k]
Since the other terms makin1! up U, 1 (j) and U, 2 ( / ) are identical, it follows that
U, 2 (j) ~ U,Jf).
r---T----
1 I I
L---- ____ _J
I I I
I I I
I I I
I
---, I
I
I
I
-+--~----~----~---x
x' x,
Figure 4.5
In case 1r2 contains several points more than 1r1 , the desired inequality
follows by repeated application of the above argument. Similarly we see that
L,t (f) ~ L,2(j).
To prove (i) we consider the partition
which consists of all the points of 1r1 and of 1r2 , in order. Then, since 1r1 < 7T
and 1r 2 < 1r, it follows from (ii) and the obvious inequality L 11 (f) ~ U,(f) that
Theorem 1 Let
A_(f) = sup L 11 (f) and A+ (f) = inf U11 (!), (4.7)

11 11
where the supremum and infimum are taken over all possible partitions 7T of
[a, b]. Then
(i) A_(j) ::;: A+(j) for every functionf;
(ii) A_ (f) = A+ (f) if and only iff is integrable over [a, b ].
In the latter case we have
A_(f) = A+(f) = A(f) =I: f(x) dx.
Proof Since L 711( / ) ::;: Uu 2( / ) for all partitions TT1 , TT2 , it follows that
L 111 ::;: inf Uu.Cf) = A+(f)

.. 2
and also that
A_(f) =sup L",(f) ::;: A+(f) .
...
Hence, for arbitrary partitions TT1 , TT2 we have
If A_(j) -=F A+(j), then (4.5) is satisfied for more than one value of A, and this
means that f is not integrable over [a, b].
Conversely suppose A_(f) = A+(f), and let A satisfy (4.5). Then
A_(f) =sup L 111(f) ::;: A ::;: inf Uu 2(f) = A+(f),

7tl
so that A = A_(f) = A+ (f). This proves that f is integrable over [a, b] and
A = f! f(x) dx = A_ (f) = A+(j). I
Let us survey the situation so far. If we are trying to determine f! f(x) dx

by using Darboux sums, we have two alternatives: we can "approximate from
above" by taking upper Darboux sums, and then take the infimum of all such
approximations; or we can "approximate from below" with lower Darboux
sums, in which case we take the supremum. The function f is then integrable,
according to our definition, if and only if these two methods lead to the same
value for f! f(x) dx.
In the next theorem we obtain another characterization of integrability,

which is useful in the ensuing theory.
Before proceeding, we recall that the equation sup.,es f(x) = oc. holds if
and only iff(x) ~ oc. for every xES and, given e > 0, there is some xES such
that f(x) > oc. - e. These properties of the supremum (and the corresponding
properties for the infimum) are used frequently in what follows.
Theorem 2 The given function f is integrable over [a, b] if and only if, for any
given e > 0, there exists a partition 1T of [a, b] such that
(4.8)
Proof Suppose f is integrable over [a, b]. Then A_(f) = A+(f). If e > 0 is
given, let 1r1 be a partition such that
and let 1r2 be a partition such that
e e
< - + A+(f) + - - A_(f) = e.
2 2
Conversely if (4.8) holds for some partition 7T, then
..1
> u..(f)- e
Since e is arbitrary, we conclude that A_(/) ~ A+ (f), and therefore (Figure

4.6) A_(f) = A+(f), so thatfis integrable over [a, b]. I
a=x~~ xt b=x.
Figure 4.6 U11(j)- L 11(f) =(shaded area).
Theorem 2 states that a function is integrable (according to the Riemann-

Darboux definition) if and only if the upper approximations U 11 (f) and the lower
approximations L 11 (f) can be made arbitrarily close to each other by choosing
1r suitably.
Corollary 1 Iff is monotonic on [a, b], then it is integrable on [a, b].
Proof Let 1r be the "uniform" partition of [a, b], in which
b-a
xk - xk-1 = -- (k = 1, 2, ... , n).
n
Assuming that f is nondecreasing, for example, we have
sup f(x) = f(xk) and inf f(x) = f(x~c- 1).

["'k-l ...k] ["'k-l"'k]
Therefore
" [f(xk) - f(xk-1)](xk -

U"(f) - L 11(f) = ~ X1c-1)
k=1
b-a
= - [f(b) - f(a)].
n
(See Figure 4.7.) By choosing n large we can therefore make U11 (f) - Lrr(f)
arbitrarily small. Hence f is integrable. I
X
a b b-a
n
Figure 4.7 U,(f)- L,(f) for increasingf.
Corollary 2 Iff is continuous on [a, b], thenfis integrable on [a, b].
Proof Let e > 0 be given. Since f is uniformly continuous on [a, b], there exists
b > 0 such that
8
lx - Yl < b implies 1/{x) - f(y)l <- -
b-a
Let 7T be any partition of [a, b] such that xk - xk_1 < b (k = I, 2, ... , n).
Then
8
sup f(x) - inf f(x) = max f(x) - min f(x) <- - ,
[O:k-l"'k] [O:k-l .,kl [O:k-loO:k] ["'k-l"'kl b- a
so that
U,(f) - L 11(f) < --

e
b- a k=1
zn
(xk - xk-1) = e.
Thereforefis integrable on [a, b]. I
Example 3
An example of an integrable function which is neither monotonic nor contin-

uous is provided by Dirichlet's function (see Figure 4.8),
.! if x = P. (in lowest terms),

f(x) = q q
{
0 if x is irrational.
124 Chopter Four Some Theorems of Calculus
Figure 4.8 Dirichlet's function.
To show that f is integrable over [0, I], consider a given e > 0. Let Q be an
integer with 1/Q < e/2 and let y 1 , y 2 , , Yn denote all rational numbers in
[0, 1] having denominators <Q. Surround each pointy; by an interval I; of
length s ef2n, and let TT be the partition consisting of the end points of all these
intervals I; . The intervals of TT then consist of (a) the intervals I; (which can be
assumed nonoverlapping), and (b) complementary intervals 1;. For the first we
have
n e e
! supf(x) (length of li) :::;; ! 1 - =-.
; It ;=1 2n 2
For the remaining intervals J1 we havef(x) :::;; 1/Q < e/2 for x in J1, so that
~ e e
~ supf(x) (length of J 1) <-!(length of J 1) :::;; - .
i JJ 2 i 2
Therefore U 11 (f) <e. On the other hand, obviously L 11 (f) = 0; this shows that
fis integrable over [0, 1].
The basic properties of the Riemann integral are listed in the following
theorem.
Theorem 3 Let f and g be integrable functions on [a, b]. Then
" (i) f (/.f(x) dx = (/. Jb f(x) dx

a a
((/.=constant);
(ii) f [f(x) + g(x)] dx = f f(x) dx + r g(x) dx;

(iii) f f(x) dx = r /(x) dx + f /(x) dx (a< c <b);
(iv) If/(x)::;;: g(x) on [a, b], then f f(x) dx ::;;: f g(x) dx;
(v) If /(x) dx I: ; : f lf(x)l dx.
We give here only the proof of part (ii); the remaining parts are outlined in
the exercises.
Proof ofpart (ii) First we use the fact that (cf. Formula (3.2), page 98)
sup [f(x) + g(x)] ::;;: supf(x) +sup g(x).

>:eS >:eS
Therefore,
n
U,(f + g)=! sup
k=l >:k-1:-:;.,:-:;.,k
[f(x) + g(x)](xk- xk_ 1)
~ U 71 ( / ) + U,(g).
Similarly,
L,(f + g) 2 L 11 ( / ) + L,(g).
To complete the proof we must use an e-argument. If e > 0 is given, there
exist partitions 1r1 and 1r2 of [a, b] such that
U,1{ / ) < r
a
j(x) dx + 2~ and U71 ,(g) < r
a
g(x) dx + 2~.
U,(f +g) ::;;: U,(f) + U,(g)

::;;: u, {f) + u,,(g)
1
: ; : f /(x) dx +f g(x) + e.
Similarly,
L 71( / + g) 2 f f(x) dx + f g(x) dx - e,

- 126 Chapter Four Some Theorems of Calculus
where TT can be assumed (by refinement if necessary) to be the same partition

as above. Since we have shown that U11 (f +g)- L 11 (f +g)< 2e, we con-
clude that J! [f(x) + g(x)] dx exists. Furthermore,
J: (f(x) + g(x)] dx ~ Uif + g) ~ J: f(x) dx + J: g(x) dx + e,

for arbitrary e, and also
J: (f(x) + g(x)] dx ;;::: L 11(f + g) ::::: f f(x) dx + f g(x) dx - e.
From these inequalities, (ii) is an immediate consequence. I
Corollary Let f be integrable over [a, b]. Then the function
is continuous on [a, b].

G(x) = r f(t) dt
Proof Let x 0 < x 1 Then by Theorem 3
IG(x1) - G(x0)1 = IJ: f(t) dt I

1
~
1
( 1/(t)l dt
where M = sup[a,bJ 1/(t)l. Since integrable functions are by definition bounded,

M is finite. Continuity of G on [a, b] is therefore proved. I
So far we ha~e only defined the integral J! f(x) dx when a <b. It is often
\}seful to consider the opposite case.
Definition 4 If a > b, we define
f f(x) dx = - Jba f(x) dx.

It is easy to verify that Theorem 3 (suitably modified in some respects)

remains valid for the case a> b. For the sake of completeness, let us also give
Riemann's original definition of the integral as the "limit of a sum." This is
the definition usually encountered in calculus texts; it does not require the use
of the supremum or infimum. It can be shown by methods like those of this
section, that the definitions of Riemann and of Darboux are equivalent.t
Definition 5 (Riemann) Let f be a bounded function defined on [a, b]. The

Riemann integral off over [a, b] is defined as follows:
provided this limit exists, and is independent of the points xk and Ek, X~c- 1 ::;;: Ek ::;;:
xk. Precisely, the limit on the right is defined to equal A if, given any e > 0, there
exists <5 > 0 such that for every partition 7T = {x0 , x 1 , , Xn} with max Axk < <5,
and for any points E1 , E2 , , En with X~c- 1 s Ek s xk, we have
Exercises
1. Letf(x) = x 2 , x e [0, I]. IfTT is the uniform partition 0 <! <

n
~n < <
I, calculate U11 ( / ) and L 11 ( / ) . Find the limits of U11 ( / ) and L 11 ( / ) as n- oo.
2. Let f(x) = [x] (the greatest-integer function). Find a partition 7T of the

interval [0, 10] with the property that U/J)- L1f(f) < 0.01.
3. Let a= x 0 < x 1 < < xn =band consider a functionf(x) which has a

constant value ck on each interval !(Xk--l xk) .. (Such a function is called a
step-function.) Find J!J(x) dx. '
4. Discuss the role of the symbol "x" in the expression J!J(x) dx.
5. Letf(x) =sin! (x =1=- 0) andf(O) = 0. Show thatfis integrable on [0, I].

X
(Hint:fis certainly integrable over [e, I] for any e > 0.)
t See W. Rudin, Principles of Mathematical Analysis, 2d ed., McGraw-Hill (1964),

Chapter 6.
6. Let f be a bounded function having a finite number of discontinuities on

[a, b). Show thatfis integrable on [a, b). (Hint: Surround the discontinuities
ofjby intervals having total length <e/2. Then use the fact thatfis con-
tinuous, hence integrable, on the complementary intervals.)
7. Fill in the details of the following proof of Theorem 3(i).
(a) Suppose ot -:::; 0. Then U,(ot/) = otU11 (f), and so on. This implies
A+(otf) = A_(otf) = ot r f(x) dx.
(b) Suppose ot = -1. Then U,(-f) = -L,(f), and so on. Consequently,
A+( -f)= A_( -f)= - J: f(x) dx.

(c) Suppose ot = -{J({J > 0). Then J! otf(x) dx = ot J! f(x) dx from
parts (a) and (b).
8. Prove Theorem 3(iv); the proof is very simple.
9. Prove Theorem 3(v). (Note thatf(x) ~ lf(x)l .)
10. Complete the details in the proof of the following lemma:
Iff is integrable on [a, b) andf(x) = Ofor x > c (where a< c <b),

then f! f(x) dx = f~ f(x) dx.
An outline of the proof is :
(a) Since f is integrable, for e > 0 there is a partition 1r of [a, b] with
c E 1r, such that U,(f) < f!J(x) dx +e. Let 1r' be the points of 7T in
[a, c). Then U,(f) = U11.(f). Consequently (with obvious notation)
A+(f; [a, c)) < f f(x) dx +e.

Similarly A_(f; [a, c))> J! f(x) dx- e for arbitrary e.
(b) Therefore A+(f; [a, c))= A_(f ; [a, c)) and the result follows .
11. Derive Theorem 3(iii) from the other parts already proved, by means of the
lemma of Exercise 10.
4.3 Integrals and Derivatives
Every calculus student knows that differentiation and integration are

inverse processes. For "indefinite" integrals this fact is included in the definition.
For "definite" integrals (for example, Riemann integrals) there are two possible
formulas:
(A) -~(r
dx a
J(t) at) = J(x);
(B) f' f'(t) dt = f(x) - f(a) .
In this section we prove these formulas under simple continuity hypotheses.

Exercise 5 of this section indicates that the continuity hypotheses cannot be
omitted.
Theorem 1 Suppose f is defined and continuous on [a, b], and let x E (a, b).
Then
!!.._(f J(t) ~t) = J(x). (4.9)

dx a
Proof From the definition of the derivative, we have
-d ix f(t) dt = lim-1[fx+h f(t) dt - Jx f(t) dtJ

dx a h-+0 h a a
= lim-1 Jx+" f(t) dt.

h-+0 h X
For fixed x we can write
f(x) = h1 Jx+h
x f(x) dt,
and therefore
.!!.._
I dx a
r f(t) dt - f(x) 1 = 1lim.!
h-+O h
r+h
x
[f(t) - f(x)] dt 1
1
~lim- IJx+h lf(t)- f(x)l dt I. (4.10)
h-+olhl x
Sincefis assumed continuous on [a, b], there exists b > 0 such that
lf(t) - f(x)l < e

(e given) when it- xi < o. Hence if ihi < o, we have
-1
ihi
II"'"'+h if(t)- f(x)i dt I :S:-
1 e ihi =e.
ihi
This shows that the limit on the right side of (4.10) is zero, and (4.9) therefore
follows. I
Example
. h(x) =
GIVen J."'' sin
- t dt, find h'(x).
0 t
Solution We apply Theorem I together with the chain rule. Thus let
u sin t
u = x 2 ; then h(x) =
0 i
- dt. Hence,
t
2
h'(x) = dh du = sin u lx = 2 sin x
du dx u x
Next we consider Formula (B).
Theorem 2 (The fundamental theorem of calculus) Let f be continuously

differentiable on an open interval (ot, {J). If ot <a < b < {J, we have
r f'(x) dx = f(b) - f(a). (4.11)
Proof For t E [a, b] define

G(t) = J: f'(x) dx.
From Theorem I we see that G' (t) = f' (t) for all t. Therefore G(t) = f(t) + ,c
for some constant C. Setting t =a, we obtain G(a) = 0 = f(a) + C, so that
C = -J(a). Hence G(t) =f(t) -f(a). If we set t = b we obtain (4.11). I
The above form of the fundamental theorem of calculus is somewhat un-

satisfactory; the basic formula (4.11) refers only to the interval [a, b], but the
hypotheses requirefto be defined on a larger interval (ot, {J) . By taking slightly
more care in the proof, we can overcome this shortcoming, as we have in the
following theorem.
Theorem 3 (Fundamental theorem of calculus: strong form) Let f be con-

tinuous on the closed interval [a, b]. Let F be defined on the open interval (a, b),
and assume that F'(x) = f(x)for a< x <b. (F is an "indefinite integral" off
on (a, b)). Then
J: f(x) dx = F(b-) - F(a+ ), (4.12)
where F(b-) = lim..,_.b- F(x) and F(a+) = lim..,__+ F(x).
Proof For a ~ t ~ b, define
G(t) = J: f(x) dx.

By Theorem l we have G'(t) = f(t) for all tin (a, b). Butf(t) = F'(t), and there-
fore G(t) = F(t) + C (C = const.) for t in (a, b). Letting t - a+, we obtain
(since G(a+) = 0 by the corollary at the end of the previous section)
F(a+) = G(a+)- C =-C.

Finally,
F(b-) = G(b-)- C
= f f(x) dx + F(a+),
and this is the same as (4.12). I
Finally, we prove the "integration by parts" formula.
Theorem 4 Let u and v be continuously differentiable on (ot, {3). Then for

ot < a < b < {3, we have
r u(x)v'(x) dx = [u(b)v(b)- u(a)v(a)]- f u'(x)v(x) dx.

Proof Let f(x) = u(x)v(x). Apply the fundamental theorem of calculus
(Theorem 2) to the function[ We leave the calculation to you. I
Exercises
1. Calculate g'(x) if
i J
2"' 2"'
(a) g(x) = e-t dt, (b) g(x) = "' e-t dt.
0
2. Show that under the same hypotheses as in Theorem 1,
-d fb f(t) dt = -f(x).
dx "'
3. Let f(x) = [x] + I and define G(x) = f~ f(t) dt. Determine G(x) ex-
plicitly for 0 ~ x ~ 2 and verify that G'(I) =I= f(I). This shows that the
hypothesis of continuity cannot be omitted from Theorem 1.
4. Iff is continuous on [a, b] and G(x) = J: f(t) dt, show (as in the proof of
Theorem I) that
n+G(a) = f(a),
. G(x)- G(a)
where n+G(a) = hm.,_a-t denotes the right-hand derivative
of G at x = a. x - a
5. Give an example of a function f which is continuous on [0, 1] and con-

tinuously differentiable on (0, I), but such that f' is unbounded on (0, I).
(Such an example destroys the hope of generalizing the fundamental
theorem of calculus too far!)
6. In elementary calculus, the "indefinite integral," f f(x) dx, isdefined to be

any function g(x) whose derivative equals f(x). Prove from this definition
that
(a) .E._
dx
Jf(x) dx = f(x), (b) Jf'(x) dx = f(x) + c.
7. On the basis of Theorem I explain why the notion of indefinite integral is
unnecessary. Why is this notion nevertheless convenient?
8. Recall that (corollary to Theorem 3 in Section 4.2) iff is integrable over

[a, b], then G(x) = J: f(t) dt is continuous on [a, b]. Under what condition
onfis it true that G'(x0 ) = f(x 0 ) for a given point x 0 E (a, b)? (cf. Exercise
3. Your answer should provide a generalization of Theorem 1.)
9. Prove the "change of variables" formula for integrals,
f f(x) dx = J: f(q;(t))q;'(t) dt,

making the following assumptions:

(i) f is continuous
(ii) cp is continuously differentiable
(iii) cp(rx) = a and cp({J) = b.
(Hint: Let F' = f and apply the fundamental theorem of calculus.)
4.4 Improper Integrals
The definition of the Riemann integral given in Section 4.2 applies only to
bounded functions f defined on bounded intervals [a, b]. There are numerous
applications, however, in which either the function or the interval becomes un-
bounded. In such cases we use the concept of an "improper integral."
Example I
t dx
Find J.
0
...;~ (See Figure 4.9).
y .
. 4.9
Fzgure J.1 dx
.J~
0
134 Chapter Fo11r Some Theorems of Calcllllls
Solution Since I/.J;;is not bounded on [0, 1], this is an "improper

integral," which is defined to be:
lim
e-+0+
rd;
yX
= lim 2(1 - .je) = 2.
-+0+
Notice that the area represented by the foregoing integral is the area of an
unbounded region. Clearly the only reasonable way to define the area of such a
region is to take the limit of the areas of regions "approaching" the given
region.
Example 2
dx
ll..,(x
eo
- = +oo.
This means, by definition, that
t dx
lim
t-++co
l --=
1
.Jx = +oo,
a fact which can easily be checked.
Definition 1 Let f be a function defined on [a, b] and such that f is integrable

on every interval [a+ e, b] withe> 0. We then define the improper integral of
f over [a, b] as
J: f(x) dx = !~~ L:J(x) dx,
provided this limit exists. If the limit does exist (finitely) we say that the improper
integral converges.
f:
An analogous definition applies to improper integrals of the form f (x) dx.
(You should supply this definition.)
Improper integrals have a theory of convergence that is very similar to the
theory of infinite series. For example, we have the following "comparison test"
for improper integrals.
Theorem I Suppose f and g are defined on [a, oo) and integrable on [a, b]for
every b > a. Suppose also that
(i) 0 -s;.f(x) -:::;. g(x) for all x ;;:::: a,

(ii) S: g(x) dx converges.
Then J;' j(x) dx also converges.
Proof We can use the simple observation that if his a nondecreasing function
defined on [a, oo ), then Iim.,__.CX> h(x) exists (finitely) if and only if his bounded on
[a, oo).
Now, fort ;;:::: a let
F(t) = f f(x) dx and G(t) = f g(x) dx.
By hypothesis (ii), limt-.ex> G(t) is finite, so that G is bounded on [a, oo). By

hypothesis (i) we have F(t) -:::;. G(t) for all t. Hence F is bounded on [a, oo) and
therefore limt-.ex> F(t) is finite. I
The following "improper" form of the fundamental theorem of calculus

is easily proved (see Exercise 4):
Lex> F'(x) dx = F(oo)- F(a),

where
F( oo) = limF(x);
Q>-+ ex>
this is valid, provided F'(x) is continuous on {ot, oo) for some ot <a and the
limit F( oo) exists. There is a similar formula for improper integrals over bounded
intervals.
As an application of improper integrals we will consider the famous
Gamma function r(x). This function is a generalization of the factorial function
{nl).
Definition 2 For x 2 I define
(4.13)
Theorem 2 The improper integral (4.13) converges for x ;;;:: 1.t Furthermore,
we have
r(x + 1) = xr(x). (4.14)
Proof We know that limt-co tae-bt = 0 for any a E ~ and any b > 0. This
implies that (with x fixed)
Since f;' e-01 2>t dt obviously converges, the integral (4.13) must also converge,
by Theorem I.
To prove (4.14), we integrate by parts:
= xr(x). I
Corollary Ifn is a positive integer, then r(n) = (n - I)!
The proof (by induction) is left for you to obtain (see Figure 4.10).
y
-+---.----.-----,--._X
2 3
Figure 4.10 r(x).
t It is easy to show that the integral (4.13) also converges for 0 < x < 1 (but not for
x .::;; 0). In this case the integral is also improper at x = 0.
Exercises
1. For which values of p does f~ xP dx converge?
2. For which values of p does f~ xP dx converge?
3. Show that the regions whose areas are represented by the following two
integrals are congruent:
4. (a) Write down the definition ofanimproperintegral oftheformf:' f(x) dx.

(b) Prove the "improper" form of the fundamental theorem of calculus
mentioned in the text (page 135).
5. Which of the following improper integrals converge?
(a) i "' -
0
x2
- dx,
1 +X 2
(b) f~ x- 312 sin x dx,
(d) J"' x(log

. dx x) 2
' (e) nlog X dx.
6. Calculate
(a) J"' -
-co
dx
-;
1 + x2
(b) f~ e-rz sin x dx.
7. Let D denote the 3-dimensional solid interior to the surface obtained by

rotating the curve y = 1/x about the x-axis, for x ~ 1. Show that D has
finite volume but infinite surface area. Is this a paradox?
8. Prove by induction that r(n) = (n - 1)! (n ~ 1).
9 .
D ISCUSS t h e Integra
. 1 Jl -dx .
-1 X
4.5 Unij01m Convergence
Let {/n} be a given sequence of functions defined on a fixed interval /.

If limn-ocJn(x) exists for each point x in/, we say that the functionf defined by
f(x) =lim fn(x) (4.15)

n-+""
is the pointwise limit of the sequence {f n} on /.

The concept of pointwise limit, which at first seems to be very natural,
suffers from several shortcomings. For example,
(a) It does not follow that the function f is continuous, even if every func-
tionfn is continuous. (See Example 1.)
(b) It does not follow thatf'(x) = limn_""f~(x), even if every functionfn
is differentiable. (See Example 2.)
(c) It does not follow that fd(x) dx =limn_"' f 1 fn(x) dx, even if this
limit exists. (See Example 3.)
Example 1
xn
Letfn(x) = + xn (0 ~ x < +oo). It is easy to verify (see Figure 4.11) that
1
f,(x) f(x)
Figure 4.11 The sequence of Example 1.
0 ifO ~ x < 1,
lim fn(x) = ! if x = 1,
n-+"" {
1 if X> 1.
Notice that although each function fn is continuous, the limit function f is

discontinuous at x = 1.
Example 2
Let gn(x) = (sin nx)in. Clearly limn-co gn(x) = 0 for every x. However, since
g;.(x) =cos nx, limn-co g~(x) does not exist. (See Figure 4.12.)
g,(x) g,(x) g,(x)
Figure 4.12 The sequenc'! of Example 2.
Example 3
Let hn(x) = nxn for 0 < x < 1. Then limn-.co hn(x) = 0 for every x in (0, 1).
However,
i l h (x) dx = n il xn dx = -n+l
0 n 0
n
-- 1 as n- oo.
This shows that (see Figure 4.13)
h,(x) h,(x) h,,(x)
Figure 4.13 The sequence of Example 3. (The areas of the shaded regions- 1.)
UO Chapter Four Some Theorems of Calculus
You will perhaps raise an objection about Example 3; it certainly seems un-
reasonable to say that the functions hn approach the function h(x) == 0 on
(0, I). Although it is true that hn(x)---+ 0 for each given point x E (0, I), the
functions hn(x) with large n do not "look" much like the zero function. To make
this more precise, we introduce the following definitions.
Definition 1 Iff and g are two bounded functions defined on an interval I, we

define the distance between f and g to be
d(f, g) = dif, g) = sup lf{x) - g(x)l. (4.16)

xei
Definition 2 Let {fn} be a sequence of functions defined on I and f another

function defined on I. If
we say that fn converges uniformly on I to f.
Returning to Example 3, notice that
d<o, 1 >(hn> 0) = sup lnxnl = n.

0<:.<1
Therefore hn does not converge uniformly to zero on (0, I). The student should
check Examples I and 2 for uniform convergence; thus for Example I, one can
compute d(f n,f) and decide whether d(f n,f)---+ 0 as n---+ oo.
To explain why d(f, g) is called the "distance" between f and g, we list
some simple properties of d(f, g):
Lemma 1 The distance d(J, g) between bounded functions f, g on a given

interval I satisfies-
(i) d(f, g)= 0 if and only iff(x) = g(x) on I,

(ii) d(f, g) 0 for all functions f, g,
:::=:::
(iii)d(f, g) = d(g,f) for all functions f, g,

(iv) d(c f, cg) = cd (/, g) for all functions f, g and constants c,
(v) d(f, g) ::::;; d(f, h)+ d(h, g) (triangle inequality) for all functions
f,g,h .
Proof Properties (i)- (iv) are trivial from (4.16). To prove (v), we have (by
Inequality (3.2), page 98)
d(f, g)= sup if(x)- g(x)l

XE[
::;; sup (if(x) - h(x)l + lh(x) - g(x)i)

O:E[
::;; sup lf(x)- h(x)i +sup lh(x)- g(x)l

O:E[ O:E[
= d(f, h)+ d(h, g). I

Notice that, graphically speaking, the distance d(j, g) represents the
"greatest" ( = supremum) vertical distance between points on the graphs of
fand g having equal abscissae (Figure 4.14).
d(f,g)
Figure 4.14
The definition of uniform convergence can be given in "e-o" terms as shown

in the following lemma.
Lemma 2 The sequence Un} converges uniformly to f on the interval I if and

only if the following condition is satisfied:
given e > 0, there exists an integer N such that,for all x E I, we have
If n(x) - f(x)l < e whenever n ::::::: N.
Proof Suppose fn ~ f uniformly on I. Then d(fmf) ~ 0. Thus if e > 0 is

given, there exists an integer N such that
sup lfix)- f(x)l = difn.J) < e for all n ::::::: N.

xel
Consequently, for every x E I,
if n(x) - f(x)i < e, if n ;:;::: N.
The converse is proved similarly. I
Example 4
Let fn(x) = xn. We can sh0w that fn--. 0 uniformly on the interval [-a, a],
provided 0 < a < I, but {f n} does not converge uniformly on [-I, I]. (See
Figure 4.I5.)
For, consider
--. 0 as n --. oo if 0 < a < 1.
This shows that fn--. 0 uniformly on [-a, a]. Since fn( -I) = ( -I)n, the
sequence {f n} does not converge at x = -I; thll;t is, the sequence{/ n} does not
converge pointwise on the interval [ -1, I]. (The sequence does converge at
x = I; we leave it to you to decide whether {fn} converges uniformly on
[0, I], for example.)
We are prepared now to prove that uniform convergence does not have the
shortcomings [(a) - (c) above] that pointwise convergence has.
-I a I
I
/,(x) f.(x) /,(x)
Figure 4.15 fn(x) = xn on [-I, I]
Theorem 1 Let {fn} be a sequence of continuous functions on an interval I.

Iffn ~ funiformly on I, then f is also continuous on I.
Proof Fix x 0 E /,and let 8 > 0. By the triangle inequality, for any x E /,
Since fn-+ f uniformly on/, we can find an integer N such that lfN(z) - f(z)! <
8/3 for all z E /. Since fN is continuous on I, there exists a ~ > 0 such that
lfN(x) - f N(x0 )! < 8/3 provided lx- x 0 ! < ~. Therefore, by (4.17), we have
!f(x) - f(x 0 )! < 8 provided lx - x 0 ! < ~.
This proves that f is continuous at x 0 I
The proof of Theorem I could be summarized as follows: since each fn

is continuous at x 0 , and since fn(x) is "arbitrarily close" to f(x) for all
x, it follows that f must be continuous at x 0
Theorem 2 Let Un} be a sequence of continuous functions on a closed, bounded

interval [a, b]. Iffn-+ funiformly on [a, b], then
J: fn(x) dx-+ J: f(x) dx as n-+ oo. (4.18)
Proof First, f is continuous on [a, b] according to Theorem I. Hence, the

functionsfn (n = I, 2, 3, ... ) and fare all integrable over [a, b].
If 8 > 0 is given, choose an integer N such that
lfn(x)- f(x)! < -8- for all x E [a, b] and all n ;;::: N.
b-a
Then
II: fnCx) dx- I: f(x) dx I S Jab lfn(x)- f(x)! dx < 8

for n;;::: N. I
Theorem 2 can be generalized as follows: if the functions fn are assumed

integrable (but not necessarily continuous) over [a, b], and fn-+ funiformly on
[a, b], thenfis also integrable on [a, b], and (4.18) holds. See Exercise 9.
Theorem 3 Let Un} be a sequence of continuously differentiable junctions

on an open interval I. Suppose that for some functions f and g on I we have
(a) fn-+ f uniformly on I, and
(b) f~ -+ g uniformly on I.
Uti Chopter Four Some Theorems of Calculus
Then f is continuously differentiable on I, and we have f' = g. Thus f~-+ f'

uniformly on I.
Proof We use Theorem 2 together with some theorems of Section 4.3. Let a
be a fixed point of I. Then for any x E /,
f' f~(t) dt =fix) - fn(a) (4.19)
by the fundamental theorem of calculus.

Letting n-+ oo and applying Theorem 2 to the left side of (4.19), we obtain
f' g(t) dt = f(x) - f(a). (4.20)
Differentiation of (4.20) yields (by Theorem I, Section 4.3)
g(x) = f'(x). I
We note that Theorem 3 becomes false if condition (b) is omitted (Exercise

7).
This completes our study of the consequences of uniform convergence. We
end this section with a famous and useful criterion that is necessary and
sufficient for uniform convergence.
Theorem 4 (Cauchy's criterion for uniform convergence) Let {fn} be a

sequence offunctions on an interval I. Then fn converges uniformly on I to some
function f if and only if
n.m-+oo
or, in other words, if and only if, given e > 0, there exists an integer N such that
lfn(x) - fm(x)l < e for all x E I and all n, m :?: N. (4.21)
Proof Suppose first thatfn-+ funiformly on/, that is, d1(fn,f)-+ 0 as n-+ oo.
By Lemma 1 (v) (the triangle inequality), we conclude that
The converse is more difficult to prove. If (4.2I) is satisfied, it follows that

for each given x E I the numerical sequence {j.,(x)} is a Cauchy sequence.
Hence (Section 3.4),f(x) = lim.,_oofn(x) exists for each x.
We must show that fn-+ f uniformly on I. Let e > 0 be given. By (4.2I)
there exists an integer N such that
lf..(x) - fm(x)l <-e for all x E I and all n, m ~ N.

2
If xis a given point of I, then, since fn(x)-+ f(x), there exists an integer N1 ~ N
such that lfN1 (x) - f(x)l < e/2.t Hence for n ~ N and for any given x E /,
we have
Thus d1 Cfn,f) :S::: e for n ~ N, and this proves that fn-+ f uniformly on I. I
Corollary Let {fn} be a sequence of continuous functions on an interval

I, and suppose Un} satisfies the "uniforYrJ Cauchy condition" (4.2I). Then Un}
converges uniformly on I to a continuous function f
This result follows by combining Theorems 4 and I.
Exercises
I. Consider the intervals
/ 1 = [0, +oo); 12 = [0, I]; 13 =[I,+oo).
For each of the following sequences of functions,
(i) find the pointwise limit, limn--+oofn(x) = f(x);

(ii) determine whether f.,-+ funiformly on each interval / 1 , / 2, and / 3
(a) f..(x) = -X ,
n
t Note that N1 may depend on x (as well as e), but this does not affect the validity of
the proof.
U6 Chapter Four Some Theorems of Calculus
2. Consider j,.(x) = nxn(l - x) on [0, I].
(a) Findf(x) = limn.... oo/n(X).

(b) Calculate d[0 , 11 (/n,f) by calculus.
(c) Doesfn-funiformly on [0, 1]?
(d) Does f!J,.(x) dx- f!f(x) dx?
3. Repeat Exercise 2 for the sequence g,.(x) = n2xn(I - x).

4. What does the statement "the infinite series !f fn converges uniformly to
fon the interval I" mean precisely?
1
5. Show that the series !:
xn converges uniformly on [-a, a] to I _ x
provided 0 <a< I. (See Example 4.)
6. Iff (x) is defined on I, then the set of all points (x, y) in the plane satisfying
ly-f(x)l < e, x EI,
is called thee-strip about the curve y = j(x).

(a) Sketch the 0.2-strip of the curve y = .x3 on [0, 1].
(b) Express the concept of uniform convergence in terms of e-strips, and
draw a picture.
7. Show by example that
fn - 0 uniformJy and /~ exists,
do not imply f~ - 0.
8. Prove that if {/n} is a sequence of functions which are uniformly con-

tinuous on an interval I and if fn - f uniformly on I, then f is also uni-
formly continuous on I.
*9. Prove that if ifn} is a sequence of integrable functions on [a, b) and if

fn- funiformly on [a, b), thenfis also integrable on [a, b) and
J: f..(x) dx- f f(x) dx.

(Hint: You wish to show that, given e > 0, there exists a parti<~on TT of
[a, b] such that U71 ( / ) - L 11 ( / ) < e.)
4.6 Power Series
Any infinite series of the form
(4.22)
is called a power series with center xo; the symbols x 0 and a0 , al> a 2 , denote
given real numbers; x denotes a real "variable."
Since there is no loss of generality in the the.o ry if we assume that x 0 = 0,
we will mainly consider the power series
(4.23)
with center 0.
Example 1
!:'=ox" is the "ge9metric series." We know that this series

(a) converges absolutelyt if.lxl < 1,
(b) diverges if lxl ~ 1.
It is a surprising fact that every power series exhibits a behavior similar to

the geometric series. Namely, given the series (4.23), there exists a number
R (0 ~ R ~ + oo ), called the radius of convergence, such that the series
(a) converges absolutely for lxl < R,
(b) diverges for lxl > R.t
We introduce the following definition of the "limit superior" of a sequence in
order to prove this fact and give a formula for R.
Definition Let {b,.} be a sequence of real numbers. We write
lim sup b,. = {J,

n-+co
t A series L:f "'"is said to conver&e absolutely if the series 2f letnl is convergent.
If +The case R = + oo signifies that the series converges absolutely for all x.
UB Chapter Four Some Theorems of Calculus
provided that the following two conditions are satisfied:

i!l Given any e > 0, there exists an integer N such that
bn < {J +e for all n ;;::: N,
(2) Given any e > 0 and any integer i, there exists an integer j ;;::: i such that
- b; > {J- e. .
If the sequence {bn} is not bounded (rom above, we define
lim sup bn = +oo;

n--.oo
finally i{lim.,_.. 00 b,. = -oo. we define
lim sup bn = -oo.
Example 2
Let bn = ( -l)n + lfn. Then

lim sup bn = 1.
n-+oo
To verify this we merely have to check that conditions (I) and (2) hold. How-
ever, these are both obvious.
Note that the sequence {bn} of this example does not converge, but it has_a
subsequence, {b 2 n}, converging to I. In general we have: for any sequence.
{b 8 }, lim sup" bn is the largest number which can be obtained as a limit of some
subsequence of {b.,}. This is easy to see from the Definition. Condition (1)
implies that no subsequence can converge to a number >{J =lim supn--.oo bn;
Condition (2) says that some subsequence does converge to fl.
Theorem I If {bn} is any given sequence, then
{J =lim sup br.

D-*00
exists (- oo < {J < + c0 and is uniquely de_termined.

4.6 Power Series U9
The proof is easy and will be omitted.
Theorem 2 (J. Hadamard) Let ~:'=o a,.x" be a given power series. Define
ex= lim sup la,.l 11". (4.24)

n-+c:o
Let R = 1/rx. (if rx. = 0, let R = + oo; if rx. = + oo, let R = 0). Then the series
~:'=o a,.x"
(a) converges absolutelyfor all lxl < R,

(b) diverges for all lxl > R.
The number_Rjs called the radius of convergence of the series~;;' a,.x".
Proof (a) Suppose first that rx. :# 0, + oo. If lxl < R = 1/rx., then for some
> 0 we have lxl < 1/rx.(l + c). By the definition we have Ja,.Jl'" < rx.(I + t:/2)
for sufficiently large n. Hence for such n,
(4.25)
where r < 1. Thus the series ~:'=o a,.x" converges absolutely by comparison
with the geometric series~;;' rn.
In the case rx. = 0 we argue as follows. Given x :# 0, by the definition we
have la,.l 11 " < 1/2 lxl for large n. Therefore lanx"l < 1/2" for large n, so that
~;;' anx" converges absolutely. The case rx. = oo is trivial.
(b) Suppose that ~:'=o anx" converges for a particular x :# 0. Then

a,.x" ~ 0 as n ~ oo, and therefore
for large n. Thus
rx. =lim sup la,.l 11" :S:: _!._,

n-+c:o lxl
so that lxl :::;:: 1/rx. = R. This proves (b). I

Example 3
The following examples are easily verified (R is the radius of convergence).
Series ot R
~n"x" +oo 0
x"
~-
n"
0 +oo
x"
~-
n
lt 1
t Recall that lim,...... ex> n1/" = 1.
It should be noted that the case Ixi = R is not covered in Theorem 2. There
is a reason for this-some series converge at lxl = R and others diverge.
Example 4
(a) The series ~;' x" has R = 1, and this series diverges for Ixi = 1, that is,
for x = + 1 and for x = -1.
(b) The series LCX) x"fn2 has R = 1, but this series converges (absolutely) for
lxl = 1.
(c) The series~:' x"fn also has R = 1. It diverges for x = +1 and converges
(conditionally) for x = -l.t
Here is another formula for the radius of convergence
I.
R =1m a ..- ,
-
n-+ex>
I I
an+l
(4.26)
provided this limit exists. Formula (4.26) is sometimes easier to use than
(4.24), but it obviously only applies to special types of power series. Since the
proof of (4.26) can be found in almost any calculus text, it is not reproduced
here.
1\t A series I: an which converges, but not absolutely, is said to converge conditionally.
Exercises
1. Find by inspection the limit superior of each of the following sequences:
1 3 1 7 1 1
(a) 1,-,-,-,-, ... , - , 2 - - , . . . ;
2 2 4 4 2n 2n
(b) {-1)";
(c) 1,2,!,4,!,8, ... ;
I
{d) (-1)" + _n_,;
n+l
(e) ~ - [~] ([ ] =greatest integer).
2. Find the radii of convergence of the following power series (use (4.24) or
(4.26) when convenient):
(c)~ (1 + ~)xn,
00 00
(a) L n2xn, (b) L 2n+lxn,
0 0
00 ..
oo n"
(d) ix:, (e) L 2n(-ll x", (f) L - x",
o n. 0 o n!
00 00
(g) L x2n, (h) LX-n.
0 0
3. (a) Define lim inf bn and discuss it.

n-.oo
(b) Find the limit inferior of the sequences of Exercise I.
4. Show that if c ~ 0, then
lim sup cbn = c lim sup bn.

n-+oo n-+oo
What if c < 0?
5. Show that, in general,
lim sup (xn + Yn) =F lim sup Xn +lim sup Yn-

n-+oo n-+oo n-+oo
152 Clulpter Fo11r Some Theorems of Calculus
What can be said in this regard?
6. If {b,.} converges, show that
lim sup b,. =limb,..

n-.oo n-+oo
7. If the coefficients in a power series are integers, and if the series is not a
polynomial, show that R :s; 1.
8. If all the coefficients in a power series are either 0 or 1, what can you say
about the radius of convergence?
9. What can you say about the radius of convergence of the series
co
L (a,.+ b,.)x"
0
in terms of the radii of convergence of the two series !; a,.x" and

!; b,.x"?
4.7 Uniform Convergence of Power Series
We prove next the following fundamental result.
Theorem 1 Let !;;' a,.x" be a given power series with radius of convergence R.
If r is any number such that 0 < r < R, then the given series converges uniformly
on the interval [ -r, r ].
Proof Since r < R = lfoc, we can write

1
r= (some e > 0).
oc(l +e)
By (4.25), if lxl :s; r, we have
where -r < 1.
We can now apply the uniform Cauchy criterion (Theorem 4, Section 4.5);
form ~ n we have (if lxl ::;:; r)
since the geometric series !~ -rk converges. Therefore, the series ! ; a~

converges uniformly on the intervallxl ::;:; r. I
Corollary The power series !; a,.x" converges to a continuous function

on the open interval (- R, R).
Proof The partial sums ! : a,.x", being polynomials, are continuous for every
x. Since these partial sums converge uniformly to f(x) = ! ; a,.x" on any
interval [ -r, r] (r < R), the sumf(x) is also continuous on [ -r, r] for every
r < R. Hence,f(x) is continuous on ( -R, R). I
Example 1
The geometric series!; x" converges uniformly on [ -r, r] (r < 1) to f(x) =

1/(1 - x); the sum is continuous on ( -1, 1) but certainly not on [ -1, 1].
(See Figure 4.16.)
-I -I
l+x I +x+x' I +x+x' +x'
Figure 4.16 Partial sums of the geometric series!; x".

Theorem 2 Let f(x) = I~ a,.x" have radius of convergence R. Then
"' f(t) dt =I-"-

a
co
f.
0
o n
x"+
+1
1
(lxl < R) (4.27)
and the power series on the right side also has radius of convergence R.
Proof Let R' denote the radius of convergence of the power series (4.27):
1' = lim sup --"-

R
a
n-+oo I n + 1
I 1
n-+oo
11
1 '" = lim sup { la,.\ "
(n
1 }
+ 1)11,.
Since Iim..-.oo (n + 1)1 '" = 1, it follows from the Lemma below that
-1 = 11m sup Ia,. 11tn = -1.

R' n-+oo R'
Now if 0 < x < R, the series I:

a,.t" converges uniformly to f(t) on the
interval [0, x]. By Theorem 2, Section 4.5, this implies that
lim
n-+oo f."' 2 akt'k dt = f."' f(t) dt.
0 k=O
n
0
But
a
=lim I __
n
k - xk+1
n-+oo k=O k + 1
The result just proved is frequently stated in the form: "any power series
can be integrated term by term within its in1erval of convergence."
Example 2
Integrating the geometric series
"' 1
Ix" = -- (lxl < 1)
o 1- X
term by term, we obtain
co xn+t
! - - =
J"' - dt = log(1- x) (lxl < 1).
on+1 01-t
Replacing x by -x, we obtain the formula
co ( on+txn
log(1 + x) = L~-~~
n=t n
x2 x3
x - 2 +3 - (lxl < 1). (4.28)
We need to prove the following lemma, which generalizes Exercise 4 of the

preceding section.
Lemma Suppose that lim Xn =ex> 0 and lim sup Yn = {J. Then
n-+co
lim sup XnYn = ex{J.

n .... co
Proof (1) Let e > 0 be given. We must show that for large n,
XnYn < ex{J + e.
Since ex> 0 we may suppose Xn > 0. Now if Xn <ex+~ and Yn < P+ y,

we have
XnYn < ex{J + exy + {J~ + ~y. (4.29)
Choosey = ef3ex and choose~ =min (e/3 lfJI, ef3y). Then (4.29) becomes
XnYn < ex{J + e (n large),

as desired.
(2) Let e > 0 and integer i be given. To show that for some integer j ;;:::: i
we have
we proceed almost exactly as above. Details are left to the reader. I

We turn next to the question of differentiation of power series.
Theorem 3 Let f(x) = !: anxn have radius of convergence R. Then

00
f'(x) =! nanxn- 1
(lxl < R), (4.30)
0
and this power series also has radius of convergence R.
Proof That the series (4.30) has radius of convergence R follows exactly as in
the previous theorem; we omit the details. Let
00
g(x) =! nanxn-l (lxl < R).

0
By Theorem 2 we have
r0
g(t) dt = i 0
anxn = f(x) (lxl < R).
Hence, by differentiation, g(x) = f'(x). I
Corollary Let f(x) = !: anxn have radius of convergence R. Then f is

infinitely differentiable on the open interval lxl < R. Moreover we have
pn>(O) = n! an (n = 0, I, 2, ... ). (4.31)
Proof This follows immediately upon repeated application of Theorem 3.
Example 3
It is perhaps surpising that the converse of the above corollary is false: a func-
tion f can be infinitely differentiable on an interval lxl < R and yet not have a
power series expansion valid on that interval. To see this, consider the function
e-11"' (x > 0),

f(x) = (4.32)
{0 (x::::; 0).
We showed in Section 2.5 that j<kl(O) = 0 for all k. If we had the expansion
f(x) =I: anxn, then we would have

an = l1 /(n)(O) = 0
n.
for all n, and this would imply that f (x) = 0. Therefore f has no power series
expansion about x = 0.
A function that has a power series expansion about x = x 0 , with positive

radius of convergence, is said to be analytic at x 0 We conclude this section by
showing that the exponential function eZ is analytic at every point; the same is
true for the trigonometric functions sin x and cos x; see Exercise 3.
The following theorem is a generalization of the mean value theorem. The
proof, which is slightly technical, is given in the next section.
Theorem 4 (Taylor's theorem with remainder) Let f ben times continuously

differentiable on the intervallx - x 0 1 < h. Then for all x in this interval we have
n-1 J(k)( )
f(x) =! ~
k!
k=O
(x - x 0t + Rn(x), (4.33)
where the "remainder" Rn(x) is given by
/(n)(~)
Rn(x) = - - (x - x 0 )",
n!
for some point ~ lying between x and x 0
Corollary Let f be infinitely differentiable on the interval lx - x0 1< h.

Then f has a power series expansion valid on this interval if and only if Rn (x) - 0
as n - oo,Jor lx- x0 1< h, where Rn(x) is given by (4.33).
Proof If Rn(x) - 0 as n --... oo, then by letting n - oo in (4.33) we obtain
oo Jlkl(x)
f(x) = ! - -0 (x - x 0)k. (4.34)
0 k!
Conversely iffhas a power series expansion about x 0 , it must have the form
(4.34). Hence letting n--... oo in (4.33) we see that limn-+oo Rn(x) = 0. I
158 Chllpter Four Some Theorems of Calculus
The series (4.34) is called the Taylor series of the function f, about the
point x 0
Example 4
e% = z-
cox"
o n!
for all x. (4.35)
To see this, substitutef(x) =~and x 0 = 0 in (4.33). SinceJ<k>(O) = 1 for

all k, we have
n-1 xk
e% = ! - + R,.(x),
k=O k!
where
e~
R,.(x) = 1 x" (0 < 1~1 < lxl).
n.
For each given x, therefore,
IR,.(x)l :::;;: el%1 lxl"--+ 0 as n--+ oo.

n!
Consequently~ has the power series expansion (4.35).
We can easily find the expansion of~ about an arbitrary point x 0 by the
following device:
e% = e% 0 e%-%o = z-n!co
co
o
(x - x0)". (4.36)
Formulas (4.35) and (4.36) are quite convenient for numerical computation of
~; see Exercises 7 and 8.
Exercises
1. Find the Taylor series expansions about x = 0 for the following functions.
Prove convergence.
(a) sin x; (b) cos x; (c) sinh x = !(~- e-%),
2. If i = J -I, "prove" Euler's formula

ei% = cos x + i sin x
by finding power series expansions for both sides. Is this a valid proof?
3. Find the Taylor series expansion of sin x about an arbitrary point x 0
4. Use the geometric series to derive explicit formulas for
co
(a) ~nx 11 ;
11=1
5. Show that the series ~f x fn
11 2
converges uniformly on the interval [ - 1, I].
6. Generalize the result of Problem 5 to arbitrary series~~ a 11xn having radius

of convergence R = 1.
7. Let x be given. Define two sequences {<5 11 } and {y11} (n = 0, 1, 2, . . .)

recursively as follows:
X
<5o= x; <511 = - - <511-1>
n+1
Yo= 1; Y11 = Y11-1 + <511-1
(b) Compute e-0 1 to 5 decimals.

(Cf. Table I of Chapter 1, page 36. Note also that the power series
for~ is alternating if x < 0.)
8. Use power-series methods to calculate J! e- 11

dt to 3 decimals.
4.8 Taylor's Theorem
We will first prove the following version of Taylor's theorem, in which the
remainder is in integral form.
Theorem 1 Let f be n times continuously differentiable on the interval

lx - x0 1< h. Then for all x in this interval we have
(4.37)
where
1
Rn(x) = J"'f< 11>(t)(x - t)n-l dt. (4.38)
(n- 1)! zo
Proof Let us use induction. For n = 1 the desired result is that iff is con-
tinuously differentiable then
f(x) = f(x 0 ) + f"'zo f'(t) dt,

a result that is very familiar.
Assume therefore that (4.37) and (4.38) are valid for n. Integrating there-
mainder Rn(x) by parts, we obtain
1
Rn(x) =
(n - 1)!
J"'"'o J<n>(t)(x - t)n-l dt
= 1 {[-/(n)(t) (x- t)n]x +.! f"' J<n+ll(t)(x- tt dt}

(n - 1)! n :x:o n :x:o
= _.!_ J<n>(xo) + Rn+l(x).

n!
Substitution of this into (4.37) proves the theorem for the case n + 1. This
completes the induction. I
In order to obtain the usual form of the remainder in Taylor's formula, we

require the following result.
Theorem 2 (Mean value theorem for integrals) Let f, g be continuous functions

on [a, b] and suppose g(x) ~ 0 on [a, b]. Then there exists a point c in [a, b]
such that
J: f(x)g(x) dx =/(c) f g(x) dx. (4.39)
Proof Let M = max[a,bJ f(x) and m = mina,bJI(x). Since g(x) ~ 0 on [a, b],
we have
mg(x) s;;,j(x)g(x) s Mg(x) (x E [a, b]),
and therefore
m J: g(x) dx s J: f(x)g(x) dx s M J: g(x) dx .

If g(x) = 0 in [a, b], the theorem is trivial. If g(x) . 0 in [a, b], then
s: g(x) dx > 0, so that we can write
f f(x)g(x) dx
s;;, m a sM.
J: g(x) dx
Now sincefis a continuous function on [a, b],f(x) must assume every value
between m and M somewhere on [a, b]. Therefore there is a point c in [a, b] such
that
f(c) =
J: f(x)g(x)dx
b '
Ia g(x)dx
and this is the same as (4.39). I
Notice that Theorem 2 is obviously also valid in the case that g(x) ::::;; 0
throughout [a, b]. We can now derive our desired formula.
Theorem 3 Let f be n times continuously differentiable on lx - x 0 1< h.

Then ( 4.37) is valid, with
(n)(c)
R,.(x) = f- -- (x- x 0 )" (4.40)
n.1
for some point c between x 0 and x.
Proof Suppose for simplicity that x > x 0 Then g(t) = (x -

t)"-1 is non-
negative fortE [x0 , x]. By Theorem 2 there exists a point c in [x 0 , x] such that
r~
f(")(t)(x - t)"-1 dt =!(")(c) r
~
(x - t)"-1 dt
= j<"l(c) (x - x 0)".
n
Hence (4.38) reduces to (4.40). The case x < x0 is similar. I
Exercises
1. Prove the binomial theorem as a consequence of Taylor's theorem.
2. Use Taylor's theorem with integral form of the remainder to evaluate the
integral J~ t"et dt.
162 Chapter Folll' Some Theorems of Calculus
3. Derive "Cauchy's" form of the remainder:
f <n>(c) 1
Rn(x) = (n _ )! (x- c)n- (x- x 0)
1
(for some c between x 0 and x) by applying the mean value theorem for
integrals with g(t) = I.
4. Show that the mean value theorem for derivatives is a consequence of the
mean value theorem for integrals.
Note: The following exercises are concerned with the interpretation of an
integral as an "average," or "mean."
5. lfjis integrable over [a, b], the number
p.(f) = -1- fb f(x) dx

b-aa
is called the mean, or average, off over [a, b]. Show that p.(f) has the
following properties:
(i) p.(c) = c if c = const.
(ii) p. is linear: p.(ocf + {Jg) = ocp.(f) + {Jp.(g) (oc, {3 constants).
(iii) f s; g implies p.(f) s; p.(g).
6. Let T(t) denote the temperature in Nome, Alaska, measured at a time t

hours after midnight of December 31, 1946. How would you define, and how
calculate, the "mean temperature" in Nome, Alaska during January 1947?
7. Explain why the name "mean value theorem" is appropriate for Theorem 2
in the case g(x) == I.
8. If g(x) ~ 0 on [a, b ], the number
J: f(x)g(x) dx
/Jg(/) = fb
g(x) dx
a
(assuming it exists) is called the weighted mean off over [a, b], with weight
function g. Show that the weighted mean off satisfies the conditions (i)-
(iii) of Exercise 5. Explain how Theorem 2 is a "mean value theorem" in
the general case. (Assume f, g continuous.)
9. Let g(x) = x and [a, b] = [0, 1]. Determine c E [0, 1] such that
f.tif) = f(c) ifj(x) = eZ.
4.9 On the Definition of the Exponential Function
In previous sections we have assumed, for the purpose of examples, that

you are familiar with the exponential function and its properties. From a strictly
logical point of view this assumption is indefensible, inasmuch as a certain
amount of "analysis" is required merely to define eZ. Fortunately we have now
come far enough to give a rigorous definition and theory of eZ-and also a"',
log x, and so forth. In fact, we could define eZ in at least three completely
different ways!
The most naive way to define eZ is first to define er for rational numbers r
(by using the method of recursion formulas of Chapter I, for example), and then
to define e"' for irrational X as limn~oo ern, where {r n} is a sequence of rational
numbers approaching x. We leave to your imagination how unpleasant it would
be to use this approach to develop the theory of the exponential function.
A second method of defining eZ is to define log x first by means of an
integral; then eZ can be defined as the inverse function of log x. This method is
quite simple and elegant, and is used in many modern calculus texts.t
The method we will use is due to Weierstrass, and is based on the elementary
theory of power series. Thus we will define e"' = Z~=o xnfn!. Historically this
is backward, but logically it is acceptable and convenient.
Definition 1 For any real number x we define
00 xn
e"'=z-. (4.41)
n=O n!
From our theory of power series we know that the series (4.41) converges
for all x. Moreover, we have, by (4.30),
d 00 nxn-1 00 xn-1 00 xn
-e"'=!-=!
dx n! n=O
-!-
(n- 1)!- nf n=1 n=O
This proves the following result.
t For example, see T. M. Apostol, Calculus, Vol. I, Blaisdell Publishing Company (1967).
Theorem 1 The function e"' is differentiable for all x, and we have
-d er:e = ez. (4.42)

dx
Notice, by the way, that if we put x = 1 in Formula (4.41), we get e1 =
I,~ 1/n!, and this is the number e, as we proved in Section 1.9. This justifies
the use of the symbol e in the definition of e"'. To justify the use of the expo-
nential notation, we should verify that for example, e2 = e e. We can do even
better.
Theorem 2 For arbitrary x, y we have
(4.43)
Proof Consider y as a constant. Define the function
From (4.42) and elementary rules of calculus, we have
Consequently f(x) =C = constant, with
C =f(O) = e11
(since e0 = 1 by (4.41)). Hence we have shown that, for any y,
(4.44)
Putting y = 0, we get e"' e-z = 1. Thus, multiplying (4.44) on both sides by

e"', we obtain finally
Corollary The function e"' is strictly increasing, and
lim er:e = 0; lim er:e = +oo.

z-+-CX> z-++G()
Proof From the definition (4.41) we see that e"' > 0 if x ~ 0. FromTheorem 2,
> 0 if x > 0; thus e"' > 0 for all x. Since
e-"' = 1/e"'
d
-e"'=e"'
dx '
we conclude that e"' is strictly increasing. Now we know that e > 2, and from
(4.43) it follows by induction that en > 2n -4- + oo as n -4- oo. The fact that e"'
is increasing shows that e"' -4- + oo as x -4- oo. Finally
lim e"' = lim e-x = lim _!_ = 0. I

~-fo-C() x-++oo a::-++oo ex
From the corollary we see that the function e"' has a uniquely determined
inverse function, defined for x > 0:
Definition 2 The function log x, x > 0, is defined to be the inverse of the

exponential function:
log x = t if and only if x = et. (4.45)
Theorem 3 We have
d 1
(a) -logx =-,
dx x
(b) log (xy) =log x + logy.
Proof (a) If log x = t, we have by calculus
dt 1 1
dx dxfdt et x
(b) Let log x =wand logy= z. Then, by (4.45),
x = ew and y = e.
Therefore,
which means that log (xy) = w + z = log x + logy. I

Definition 3 Let a > 0 be given. Define

a"'= e:toga. (4.46)
The usual properties of a"' can immediately be deduced from (4.46); see
Exercise 1.
Exercises
I. Prove the following formulas. (x, y, ... denote arbitrary real numbers,
whereas a, b, .. . denote positive real numbers.)
(a) (ab)"' = a"'b"', (b) log a"'= x log a,

d
(c) (a"'Y' =a""', (c) dx a"' = a"' log a,
rxdt
2. Show that log x = f1 t
-.
3. If a > o, show that

Xa
log x <- (x ;;;::: 1).
a
(Hint: Use Exercise 2.)
*4. If a ;;;::: 0, b ;;;::: 0, show that
(a+ b)"' ::;: a"' + b"' (0 ::;: x ::;: 1).
The following exercises indicate how the properties of the trigonometric

functions can be derived from the power series for these functions.t Thus
we define the functions s(x) and c(x) as follows.
xa x5 x2n+1
s(x) =X--+--+ (-1)n
3! 5! (2n+1)!
+ '
x2 x4 x2n
c(x) = 1 - - + - - + (-1)n- + .
2! 4! (2n)!
t For a complete treatment, see K. Knopp, Theory and Application of Infinite Series,
Hafner (1951).
5. Verify that s'(x) = c(x) and c'(x) = -s(x).

6. Show that
s(x + a) = s(x)c(a) + c(x)s(a),
c(x + a) = c(x)c(a) - s(x)s(a).
(Hints: (a) Let f(x) = s(x +a) - s(x)c(a) - c(x)s(a). Show that f" +
J= 0. (b) Hence show that [(/')2 + f2]' = 0, so that (/') 2 + f2 = const. =
0. (c) Deduce the above formulas.)
7. Show that (s(x))2 + (c(x))2 = 1.

5 Limits and Continuity in n Dimensions
In this chapter we wish to generalize the theory of limits and continuity

to deal with functions of several variables. For the most part, the generalization
is very natural and simple. It depends primarily on the observation that limits
are defined in terms of distances between points: for example, the inequality
lx - al < 15 is a statement about the distance between x and a. The definition
of distance generalizes easily to n dimensions.
In discussing functions of one variable, we invariably considered functions
defined on an interval /. Although the notion of an interval generalizes easily
to n dimensions (for example, a rectangular "box"), it is obvious that if our
theory is going to be useful we must consider a much wider class of n-dimen-
sional objects than just "intervals." Not only do we wish to consider arbitrarily
shaped "solid" regions, but we are also interested in more complicated objects,
such as "surfaces," or "curves." Such objects have no interesting counterpart
when n =I.
A major complication, then, in passing from one dimension ton dimen-
sions, lies in the nature of the subsets of n-dimensional space that we consider.
Fortunately this leads to only relatively minor complications in the study of
limits and continuity. It leads to tremendous additional complications, how-
ever, in the study of the calculus itself. These complications, treated in texts on
advanced calculus, vector analysis, differential geometry, and so on, are beyond
the scope of this text.
168
5.1 Open and Closed Sets in ytn 169
5.1 Open and Closed Sets in r!Jn
A point in Euclidean n-space is, by definition, an ordered n-tuple of real

numbers:t
(5.1)
The number x; is called the jth coordinate of x. The set of all such points is
denoted by r!Jn, and r!Jn is called Euclidean n-space.
It is convenient also to introduce a "linear structure" on r!Jn by defining
addition and scalar multiplication as follows:
(5.3)
Geometrically speaking, it is useful to visualize the points x in r!Jn either

as points, or as vectors pointing from the origin 0 = (0, 0, . .. , 0) to x (Figure
5.1). If we use the vector interpretation, then Equations (5.2) and (5.3), re-
spectively, become the usual parallelogram law and scalar multiplication law
for vectors (see Figure 5.2).
(x' , x',x' )
i'Thepointx
I
The vector x I
I
I
I
/L---1-----,--~ x'
x'
Figure 5.1 Points vs. vectors in r!Jn(n = 3).
t We use the superscript notation for coordinates and reserve subscripts for use with
sequences.
170 Chapter Five Limits and Continuity in n Dimensions
ax.,.-' i
...... I
I I
I I
I I
I
I /
I /
I /
~o/
(a) (b)
(a) Parallelogram law of addition (b) Scalar multiplication
Figure 5.2 Vector operations (n = 3).
The following rules of calculation are immediately obvious from the

definitions of Formulas (5.2) and (5.3).
Properties of vector operations in !Jin

1. X+ y =y + X.
2. x + + (y z) = (x + y) + z.
3. oc({Jx) = (oc{J)x.
4. oc(x + y) = ocx + ocy.
5. (oc + {J)x = ocx + {Jx.
6. x -x=O.
7. Ox= 0.
Definition 1 If x E !Jin, we define the length of x as
(5.4)
If x, y E !Jin, we define the distance between x andy as
(5.5)
It is of basic importance that this distance satisfies the triangle inequality.

To prove this we need the following lemma.
5.1 Open and Closed Sets in &~n 171
Lemma I (Schwarz's inequality) For any x, y E [Jln,
(5.6)
Proof Let x, y be given, and let oc denote an arbitrary real parameter. Then
n
0 S:: l x + ocy ll 2 = ! (xi + ocl) 2
i ~l
n n n
=! (xi)2 + 2ot! xiyi + ot2! (/)2
i~l i~l i~l
Now by elementary algebra we know that if the quadratic expression A +

2Boc + Coc2 is ::2::0 for all values of oc, then B2 - AC :::;: 0. Therefore,
This is exactly equivalent to (5.6). I
Definition 2 If x andy are vectors in [Jln, the expression
n
x y = !xiyi
i~ l
is called the dot product (or inner product) of x andy.
Schwarz's inequality can therefore be written as
lx yl s ll x ll lly ll .
Angles can also be introduced into [Jln by means of the formula
COS 8 = X y
ll x ii iiY II '
which determines a unique angle 8, 0 :::;: 8 s TT, called the angle between the
vectors x andy.
Theorem (Triangle inequality) For any x, y E gen
llx + yll ~ llxll + llyll . (5.7)
Proof Consider
nx + yll 2= ! (xi + yi)2

= ! (xi)2 + 2! xiyi +! (yi)2
~ llxll 2+ 2ll xll llyll + llyll 2 (by Schwarz's inequality)
= (llxll + l yll )2. I

Corollary For any 3 points x, y, z in &en,
l x - yll ~ ll x - zll + li z - yll . (5.8)

l\ u- -z) -\ c-:r.- '-)) \i
The pro of is obvious. Geometrically the inequality (5.8) is indeed a fact
about triangles, see Figure 5.3 .
Figure 5.3
Definition 3 Let x 0 be a given point of ~n and let r > 0. Then the set N~(x 0 ),
defined by
(5.9)
is called the (open) ball of radius r centered at x 0 The set S.(x0 ), defined by
(5.10)
is called the sphere of radius r centered at x 0

The ball Nr(x 0 ) is also sometimes called the r-neighborhood of x 0
5.1 Open and Closed Sets in /Jt" 173
Definition 4 Let D be a given subset of~". We say that D is an open set

provided that every point x of Dis the center of some ball Nr(x), r > 0, contained
within D.
The whole space D = ~" is obviously an open set; so is the empty set,
0 (since 0 has no points, the condition of Definition 3 is surely satisfied!). 1, u
Another easy example of an open set is the half-space '' -&~ '' l~vV...J.M vv<rl"
H = {x E I
~" x 1 > 0}.
Lemma 2 The ball N,(x 0 ) is always an open set.
Proof Let z 0 E Nr(x 0 ). Then Jl x0 - z0 ll = p < r. We can show (Figure 5.4) that
Nr_v(z0 ) is contained in Nr(x0 ). For if x E Nr_'P(z 0 ), then l x - z0 l < r - p, so
z.
Figure 5.4
that, by the triangle inequality,
Jl x - Xoll :S: l x - zoll + l zo - Xoll < r - p +p = r,
that is, x E Nr(x0 ). This proves that Nr(x0 ) is open. I
Notice that when n = 1, the ball Nr(x0 ) is the same as the open interval
(x 0 - r, x 0 + r).
Definition 5 A subset D of ~n is called closed if the "complementary set"

D' = {x E ~n I x '. D} is an open set.
Obvious examples of closed sets are !!4" itself (whose complement is the
open set 0 ), the empty set, 0 (whose complement is ~"), and the closed half-
space
H = {x E !!4" I x 1 ::2': 0}.
171 Chapter Five Limits and Contiflllity in n Dimensions
Another example is the closed ball
(see Exercise 3). In particular, a closed inter1al [a, b] in f!l is a closed set.
Let us point out that a given subset of fJln need not be either open or closed.
For example, a half-open interval (a, b] in f!l fits neither definition.
One further notion which is useful in studying continuity is that of the
boundary of a set.
Definition 6 Let D be a subset of fJln and let x 0 E fJln. Then x 0 is called a

boundary point of fJln provided every ball N,(x 0 ) contains some point of D and
some point of the complement, D' (see Figure 5.5). By the boundary of D we
mean the set of boundary points of D.
A boundary
pointofD
Figure 5.5
Example
The boundary of the ball Nr(z 0 ) is the sphere Sr(z0 ). This is intuitively obvious
and is easily verified by considering the straight line L through z0 and x 0 E Sr(z0 ).
Any ball N'P(x 0 ) contains points on L both inside and outside Nr(z0 ) (Figure
5.6).
Figure 5.6
5.1 Open and Closed Sets in EJln 175
Exercises
I. Let x = {3, 1, 0) and y = (1, -1, 2) be points in PA3 Calculate llxll,

llyll, llx + yll, and .! xiyi, and check that the Schwarz and triangle in-
equalities are valid.
2. Prove the inequality
(~/J ~ n i~ (xi) 2

3. Show that Nr(x0 ) = {x E Blnlllx- x0 ll ~ r} is a closed set.
4. Decide whether the following sets are open or closed or neither.
(a) {x E PA 3 Ix 1 = 0};
(b) A set consisting of a single point;
(c) {x E fAn I X =F 0};
(d) {x EPA I x = 1/k for some integer k = 1, 2, 3, ... };
(e) {x E PA2 1 x 1 is rational};
(f) N(O) n ii = {x E PAn illxll ~ r and x 1 ::?: 0}.
5. Prove that the intersection of two open sets D 1 and D 2 is an open set.
6. Prove that the intersection of a finite number of open sets is an open set.
7. Show that the intersection of an infinite number of open sets may not be an
open set.
8. Prove that the union of any number (finite or infinite) of open sets is an open
set.
9. I
Prove in detail that {x E PAn x 1 = 0} is closed.
10. Show that a set D is closed if and only if it contains all its boundary points.
11. Show that the union of a set D with the set of its boundary points is a closed
set. (This set is called the closure l> of D.)
12. Show that

(a) the union of finitely many closed sets is a closed set,
(b) the intersection of an arbitrary family of closed sets is a closed set.
(Use the results of Exercises 6 and 8.)
5.2 Sequences and Limits
The limit of a sequence {x;} of points in !Jtn is defined by an obvious

adjustment to the definition of Chapter 1.
Definition Let {x;} be a sequence of points in (f.ln, and let x 0 E !Jtn. We say
that {x1} converges to x 0 ; symbolically,
lim X;= x 0 [or X;--... x 0 asj--... oo],

;~oo
provided the following condition is satisfied:

given e > 0, there exists an integer N such that
- jjx;- x0 j < e for allj ~ N. (5.11)
Notice that lim 1_ 00 X; = x 0 if and only if lim;-oo llx; - x0 11 = 0.
Lemma 1 The sequence {x;} converges to x 0 if and only if each of the coordinate
sequences {~}f::, 1 converges to x'O,for k = 1, 2, ... , n. -'> : -xj 1
x~ -1 -./-..
J ")
Proof First, if X;--... x 0, then for each k = 1, 2, ... , n, 1 .J
:>t; -: J(>
= llx;- Xoll--... 0 as j--... oo,
so that ~--... x'O. Conversely, if~- x'O for each k = 1, 2, .. . , n, then
as j--... co,
so that X; - X 0 I
The reason for the adjective "closed" in the term "closed set" is explained
by the following simple result.
Lemma 2 Let D be a subset of~n. Then Dis closed if and only if every con-
vergent sequence of points in D converges to a point of D.
( ::-2))
Proof Suppose D is closed. Let {x1} be a sequence of points of D, with
lim 1_ 00 x 1 = x 0 If x 0 = D, then x 0 ED'. Since D' is open, there exists a ball
f Nr(x()), r > 0, contained in D'. But each x 1 is in D, so that llx1 - x 0 11 ~ r,
which contradicts the assumption that x 1 ~ x 0 Therefore x 0 ED.
Conversely, assume that every convergent sequence of points in D con-
verges to a point of D. Suppose that Dis not closed; in other words, D' is not
open. Then there is a point x 0 in D' such that no ball Nr(x0 ), r > 0, is contained
in D'. Thus every ball N 111 (x0), j = 1, 2, 3, ... , contains some point x 1 of D.
Since JJx1 - x0 JJ < 1/j, the sequence {x1} converges to x 0 =D. This is a contra-
diction, and therefore D must be closed. I
Example
We can easily check that the set
is closed by using Lemma 2. For if {x,} is a sequence in n with x, ~ Xo, then

x} ~ 0 for all j, so that (by Lemma 1) x~ = lim 1_ 00 x} ~ 0. This means that
Xo E 0, and therefore R must be closed.
The following simple properties of limits of sequences in ~n can be easily

proved.
Theorem 1
(a) If {x1} converges, the limit is unique.

(b) If {x1} converges and~ is a constant, then
lim ~x 1 =~lim x 1
i-+oo i-+oo
(c) If {x1} and {y1} converge, then
lim (x 1 + y 1) =lim x 1 +lim y 1

i-+oo j-+oo i-+oo
178 Chllpter Five Limits and Continuity in n Dimensions
Next we come to the Cauchy criterion for convergence. A sequence {x1}

of points of !Jln is called a Cauchy sequence provided that:
given s > 0, there exists an integer N such that
//x1 - xk/1 < s for allj, k 2: N. (5.12)
The condition (5.12) can also be expressed by writing
lim l/x 1 - xk/1 = 0. (5.13)

i ,k-+ 00
Theorem 2 (Cauchy's criterion) A given sequence {x1} in !Jln converges if and

only if it is a Cauchy sequence.
Proof Suppose {x1} is a Cauchy sequence. Then, as in Lemma 1, each co-

ordinate sequence {xn, I = 1, 2, ... , n is also a Cauchy sequence, since
/x}- xi/ $;; /IX;- xkll Hence, Iim;-.oo x} = xJ exists for each/. By Lemma 1,
Conversely, if Iim 1.... 00 x 1 = x 0 , then each coordinate sequence converges,

and is therefore a Cauchy sequence:
lim /x} - x;/ = 0 (l = 1, 2, ... , n).

i,k-+oo
Consequently also
as j, k~ oo;
that is, {x;} is a Cauchy sequence. I
We prove next the n-dimensional version of the Bolzano-Weierstrass

theorem. If D is a given subset of !Jln, we call a point x 0 in !Jln an accumulation
point of D if every ball Nr(x 0 ), r > 0, contains some point of D not equal to
x 0 It is clear that x 0 is an accumulation point of D if and only if there is a
sequence {x;} of points of D which converges to x 0
Theorem 3 (Bolzano-Weierstrass) Every infinite, bounded subset of !Jln has an

accumulation point in !Jln.
"I 'dA ''/ y

VV'qc-
Proof If Dis bounded, it is contained in some lcube Q0 of side r0 Divide Q0
into 2n congruent cubes of side (l/2)r 0 , by bisecting each edge of Q0 Let Q1
/:<
denote one such cube containing infinitely many points of D . L.X., o.. ( '? t~-., -.--l ~~
Starting with Q1 , we obtain by the same process a sub-cube Q2 of side
ir containing infinitely many points of D, and so on. Let {x;} be a sequence of
0,
points of D with X; E Q; for all j. Then {x1j is a Cauchy sequence, because if
j, k ::2: N, then x 1 and xk belong to QN, so that
for sufficiently large N. By Theorem 2 the sequence {x;} converges to some point
Xoin (jln , I
Corollary Every bounded sequence in (jln has a convergent subsequence.
Exercises
1. Let X; --
---
( - .j - - ,
J + 1 31
1) 11112
Eon (j = 1' 2, 3, 0 0 .).
(a) Find lim;__. 00 X; = x 0

(b) Given e > 0, determine an integer N such that
llx; - x 0 ll <e for all j ::2: N.
2. Prove Theorem 1(c).
3. Given lim 1__. 00 X; = x 0 and lim;__. y 1 = y 0 , prove that

00
lim X; y1 = X0 Yo
i-+ 00
4. If X; = (sin (frr/2), cos (jTT/2)), show that {x;} is bounded and find a
convergent subsequence.
5. Find the set of accumulation points of the sets of Exercise 4, Section 5.1.
6. Express the statement "D is a closed set" in terms of the accumulation

points of D, and prove your statement.
7. Show that if lim 1__. 00 X; = x 0 , then
llx;- zoll-+llxo- zoll asj-+ oo
for any fixed point z 0 (Hint: Use the "backward" triangle inequality in
~n.)
8. (a) If {x;} is a given sequence in ~n, define ~~ 1 X;.
(b) Prove that ~~ 1 X; converges if ~1'= 1 /lx;/1 does so. (Use the Cauchy
criterion!)
(c) Prove or disprove the converse to (b).
9. A set S in ~n is said to be dense if every nonempty open set D in ~n
contains a point of S.
(a) Show that Sis dense if and only if every point of ~n is an accumulation
point of S.
(b) Show that there is a countable, dense subset of ~n.
5.3 Limits and Continuity
In advanced calculus it is customary to begin by studying real-valued

functions f (x1 , x 2 , , xn) of several variables. In this text we immediately go
one step further and discuss vector-valued functions of several variables. This
means that the values
will themselves be vectorst (or points) in another space ~m:
j(x) = (jl(x),f2 (x), ... ,fm(x)).

Example 1
Consider the functionf = (f1 , f2, f 3

) determined by the equations ~,
--:' -:: jl(O, cp) = sin() sin cp,} '

(:x , ?' l ;;'') ~ -y e !<-"
x' :.:. / 2 (0, cp) = sin() cos cp, (5.14)
) ' ' :: /
3
(0, cp) = cos e.
Thenfis a function defined on ~2 , with values in ~3 We symbolize this fact
by writingf: ~ 2 -. ~ 3
t In many texts specia! notations, such as arrows, bars, or boldface type, are used for
vectors. Such notations, although sometimes convenient, are not necessary if the context is
clearly understood.
Definition I Let D be a given subset of !Jn. We use the notation
to indicate that f is a function defined on D and having its values in !Jm. The set
D (which henceforth will always be specified) is called the domain of the function f
The set of values off, namely {f(x) I x E D}, is called the image of D under f and
is denoted by f[D].
The excuse for introducing such general functions at the outset is, first,
that questions about limits and continuity are as easily handled in the general
as in the specific case, and second, that there are many useful applications.
Definition 2 Let D be a subset of !Jn and let f: D ~ !Jm. If x 0 is an accum-

ulation point of D, we say that
lim f(x) = a
:.e~zo
provided the following condition is satisfied:

for every e > 0 there exists b > 0 such that
ll f(x)- a ll < e whenever xED and 0 < ll x - x 0 l <b. (5.15)
Notice the similarity of this definition with the definition of limit for an
ordinary function of one variable (Section 2.2). There is one subtle difference,
however. Namely, we are not assuming thatfis defined in some whole neigh-
borhood Nr(x 0 ). Hence, we have added the stipulation "x E D" to the condition
(5.15).
Recall the numerical practice you had in Chapter 2 of calculating b for
given e. Here is a simple 2-dimensional example for you to practice on (don't
be discouraged if it turns out to be tricker than it looks!).
Example 2
Let f: !J 2 ~ fJ. 1 be given by

f(x, y) = (x + 2y) 2

Obviously lim<x.vl--+<a.2 > f(x, y) = 49. Given e > 0, determine b so that (5.15)
holds.
The following theorem ought to be patently obvious, and we state it

only roughly.
Theorem 1
(a) Limits are unique if they exist.

(b) lim (f(x) + g(x)) =lim f(x) +lim g(x), if these exist.
(c) limcf(x) = climf(x), if this exists.

z-+zo a:-+:z:o
We can now define continuity in terms of limits.
Definition 3 Let f: D -+ ~m and let x 0 E D. Then f is said to be continuous at

Xo if
lim f(x) = f(xo).
z-+zo
Furthermore,fis said to be continuous on D if it is continuous at each point of D.
Theorem 2 Let f, g: D -+ ~m, and let x 0 E D. Iff and g are continuous at x 0 ,

then so are the functions f + g, rxf ( rx = constant) and f g.t
Theorem 3 Let f: D-+ ~m and let g: E-+ ~k, where we suppose that the
image off is contained in E. Iff is continuous at x 0 E D and g is continuous at
y 0 = f(x 0 ), then the composite function g of: D-+ ~k is continuous at x 0
Proof(cf. Section 2.4) Let e > 0. Then there exists 151 > 0 such that
llg(y)- g(y0)11 < e for ally E E with lly- Yoll < 151.
Also, there exists 15 2 > 0 such that
llf(x) - f(x 0)11 < 151J for all xED with llx - Xoll < 152.
'-
Hence, xED with llx- x 0 ll < ~ 2 implies that
llg(f(x)) - g(f(xo))ll < e. I
t Note that f g: D- B1
The following result states that addition and multiplication are continuous
operations.
<. ;J I
Theorem 4 The functions f -z_ . /<.. -7/'-
h(x,y) = x + y, fix,y) = xy
are continuous as functions of two variables (that is, as functions defined in &? 2).
&- ~ ll-"') : ~) : "'. -,r<.l
Proof Let ~(x, y) = x and 'Y(x, y) = y. It is obvious from the definition that
~ and 'Yare everywhere continuous. Since f 1 = ~ 'Y and f 2 = ~ 'Y, Theorem +
2 implies that h and h. are continuous. I
oM- ~U ,, -=. t,", ( .. ,;(0 cP(Yu; &;u)
Example 3 ('00H (X<,) '1o) \ J' ,li j-) '. "1 1.)
Consider the function
f(x, y) = 2~ 2 _(x2 + y2 =I= 0).

X y
. Obviously this function is continuous everywhere except (0, 0), where it is not
defined. To study the behavior off near the origin, we will show that
((II II) I)
(i) lim f(x, y) does not exist, although
(o:.v)-+(O.O)
(ii) f(x, y) is bounded; in fact lf(x, y)l <!for all x, y.
Figure 5.7 The surface z = xy

X2 + y 2
The proof of (ii) is easy : use the elementary inequality 2 lxyl :::;:; x 2 +y 2

To check (i), consider points (x , ax), where a = const. We have
a
f(x, ax) = - - 2 ,
1 +a
which is constant (see Figure 5.7). Now ax---.. 0 as x---.. 0, so that if
lim f(x,y)
(X.Y )--> (0.0)
exists, it must equal

limf(x, ax) = _a_
x->O 1 +a 2
for any a. But this value varies with a; for example, it equals 0 if a= 0 and 1/2
if a = 1. Hence (i) is proved.
The definition of continuity used so far has been in terms of limits. However
there is another important characterization of continuity in terms of "inverse
i!llages" of open sets, which we give now. Given the function f, by the inverse
image of a set U, we mean the set
J- 1 [U] = {x E domain ofJif(x) E U}.
Theorem 5 Let D be an open set in f!Jtn and let f : D ---.. f!Jtm . Then f is con-
tinuous on D if and only if the inverse image f -l[ U] of every open set U in f!ltm
is an open set in f!Jtn.
Proof First, suppose fis continuous on D. Let U be an open set in f!Jtm and let
x 0 EJ-1 [U]. Then y 0 = j(x0 ) E U, and since U is open, there is a neighborhood
N,(y0 ) c U. Since f is continuous at x 0 , there is a number o > 0 such that
llf(x)- j(x0 )11 < e, provided llx- x 0 ll < o. This implies that
~x 0 ) c J-1 [N,(y0)] c J-1 [U],

and hence we have shown thatj-1 [U] is open.
Conversely, suppose J-1 [U] is open for every open set U c f!Jtm. Let
x 0 ED and Iety0 = j(x0 ) . Thenj-1 [N.(y0)] is open, for any e > 0, so that there
exists some neighborhood N~(x0) c J-1 [N,(y0)]. This implies that
llf(x) - Yoll = llf(x) - f(xo)ll < e
whenever llx- x 0 ll < o. Thusfis continuous at any point x0 ED. I

Example 4 \'\:.. 1
1 ~
X~ \R~} (~ (u~.J
1
D- "' I'<': 1 \hm t.{-;;.) k S" 6r U: lZ
Letf(x,y) = x 2 + 4y 2
be defined on ~ 2 Then the inequality ,, ~('-"..-
_, /--~} 1 - ~
X
2 + 4y 2
<
4
-'-V f::\, ~-~r\---,-,rt
- L-,-,-+ -1 - /2 ;.r
determines an open set in ~ 2 , since this set is just the inverse image of the open
interval (-oo, 4). -+~ ' (_
~ 1 q) ~ i_ {_'X 1'1) e lt1 .f:;;(J'/'1) e (-o?; l.!)f ~;C) l
It can also be easily shown that the inequality
x2 + 4y2 ~ 4
'
(11\
(
.e \'; ' <" (A)IV /)
l
-cn-t ..-l ! ) O'f/'1'
determines a closed set in ~ 2 !'A similar sit:~ation prevails for any continuous
functionf:~n-+ ~.
We conclude this section with some examples of functions from ~n to
~m, m 9= 1.
Definition 4 Let xl> x 2 be points of ~m and let I= [IX, p] c ~ 1 Any con-

tinuous function f: I-+~"' such that f(iX) = x 1 and f(P) = x 2 is called a curve in
~m,joining x 1 and x 2.t
Example 5
I }
J-Gc !?, 1 t : 14 fR I h"'
Consider the curve f: [0, 21T] -+ ~ 3
given by
jl(t) =cost,
f2(t) = sin t,
f3(t) = t.
Tnis is a helix lying on the cylinder (x1) 2 + (x = 2 2

) 1 and joining the points
(1, 0, 0) and (l, 0, 21r). (See Figure 5.8.)
t ":-1) t -- l..-<1'
t This is the modern definition of a curve as a continuous function. It may disagree

slightly with the intuitive notion of a curve as a set of points, which now becomes the image
of the curve.
186 Chopter Five Limits and Conti1111ity in n Dimensions
x'
'
D .' s
/
I
-
X'
x'
Figure 5.8 A helix.
Definition 5 Let D be an open set in rJe2, with boundary S. A continuous

functionf:(D US)- Pln is called a (two-dimensional) surface in Pln.
Example 6
\o1 \<t-
b = [ o 1 l1T1;x::[o 11f1
Let D be the rectangle 0 S: () S: 21r, 0 s; cp ~ 1r. Consider the function
f: (D uS)- Pl3 given by Equations (5.14). Notice that
so that all points "on" the surface f lie on the unit sphere in Pl3 (Figure 5.9).
x'
7T
-
()
f
27T x'
Figure 5.9 A spherical map.

Conversely, by trigonometry one can easily see that every point on the uriit
sphere is the image of some point (0, c/>) of D u S. The function fin fact could
be used to construct a map of the globe (but most actual maps utilize more
complicated functions than this). The coordinates (0, c/>) associated with a given
point x, when expressed in degrees, are the longitude and colatitude of x,
respectively.
The functions of the last two examples are not only continuous but also
infinitely differentiable, .since the trigonometric functions used in their con-
struction are infinitely differentiable. In the early days <>f calculus, it was more
or less tacitly assumed that all continuous functions were infinitely differentiable.
Of course the function lxl is a simple counterexample to this assumption.
Weierstrass jolted the mathematical world in 1872 when he published his ex-
ample of a continuous function f: Bl __.. Bl which was nowhere differentiable.
Soon after, G. Peano (1858-1932) administered another shock with his "area-
filling curve," consisting of a continuous functionf: [0, 1] __.. Bl2 , whose image
is an entire square. Modern versions of these examples are given in Section 5.5
below:
Exercises
1. Let f(x 1 , x 2) = lx1 + x 2 1. Given x 0 e Bl2 and e > 0, determine <5 >0
such that 1/(x) - f(x 0)1 < e for llx - x0 ll < <5.
2. Letf(t) = (tcos t, tsint), so thatJ:fA1 _.. PA 2 Given e > 0 determine

<5 > 0 such that 11/(t)- /(77)11 < e provided It- 171 < <5.
3. Describe the following space curves geometrically:

(a) f(t) = (t cost, t sin t, t) (0 :::;:: t :::;:: 477);
(b) f(t) = (cost, sin t, sin t) (0 :::;:: t :::;:: 277);
(c) f(t) = (t, 2t, 3t) (- oo < t < oo).
4. Describe the following surface geometrically:
jl(O, c/>) = (2 +cos c/>) cos 0,

f"'(O, c/>) = (2 +cos c/>) sin 0,
/ {0, c/>) = sin cf>,
3
where 0 :::;:: 0, c/> :::;:: 217. Sketch the surface and its "coordinate curves"
0 = const., and cf> ~ cons~.
5. If x 1 , x 2 are given points of !Jin, find a curvef: [0, 1]-+ !Jin which corre-
sponds to the straight line segment from x 1 to x 2
6. Letf= (f\f 2 , ,fm) be a function from D c !Jin to !Ji"'. Prove that

f is continuous at x 0 ED if and only if eachji:D-+ fJi (j = 1, 2, ... , m)
is so.
7. Letf: !Jin-+ fJi be a continuous function on !Jin. Show that for any con-
stant c,
(a) {x E !Jin if(x) < c} is open,

(b) {x E !Jin IJ(x) ~ c} is closed,
(c) {x E !Jin if(x) = c} is closed.
8. Describe the following sets geometrically, and determine whether they are
open, closed, or neither.
(a) {(x,y) E &i 2 j xy > 1},
(b) {(x,y)E&i 2 jx<y ~x+ 1},
(c) {(x, y, z) E &i 3 j x + y + z ~ 1 and x 2 0, y 2 0, z 2 0}.
9. Show that the image of a curve f: [a, b]-+ !Jin is a bounded set in !Jin.
(See Exercise 6.)
10. Prove that "division is a continuous operation" with the obvious proviso.
5.4 Properties of Continuous Functions
In this section we generalize to n dimensions the theorem that a con-

tinuous function defined on a closed, bounded interval is bounded and assumes
minimum and maximum values (cf. Section 3.5). We also consider uniform
continuity and uniform convergence.
I ,p/JY"
<; O't
Definition A subset S of !Jin which is both closed~nd bounded is said to be

compact.
Theorem I A set D c !Jin is compact if and only if every sequence in D

contains a subsequence converging to a point of D.
Proof Suppose first that D is compact. Then any sequence {x1} in Dis bounded
and must have a convergent subsequence {x1.} by the Bolzano-Weierstrass
theorem. Since D is closed, the limit of this subsequence is a point of D.
Conversely, suppose every sequence in D has a subsequence converging to
some point of D. In particular, given x 1 ED, if x 1 -+ x 0 asj-+ oo, then x 0 ED;
this shows that D is closed. If D is unbounded, then there is a sequence {x1} in
D with ll x1 11 -+ + oo. Clearly {x1} can have no convergent subsequence. There-
fore D is also bounded. I
The next theorem generalizes the fact that a continuous function on a closed,
bounded interval is bounded, to the case of a continuous functionf: D-+ tnm.
Theorem 2 Let D be a compact set in tnn, and suppose f :D-+ tnm is a

continuous function. Then the image f[D] is also a compact set.
Proof Suppose first that f[D] is not a bounded set. Then there must exist a
sequence {x1} in D such that llf(x1)11 -+ oo as j-+ oo. Let {x1.} be a subsequence
of {x1} converging to a point x 0 E D. Since f is continuous at x 0 , we have
f(x;k)-+ f(x 0 ) ask-+ oo, so that
From this contradiction we conclude thatf[D] must be bounded.

To show thatf[D] is closed , we show that if {z1} is any sequence inf(D],
converging to some point z 0 , then z 0 Ef[D]. To see this, note that z 1 = f(x 1) for
some x 1 ED. The sequence {x1} has a subsequence {x;.} with limk-oo x 1k = x 0 ED.
Hence, z 1k =f(x 1)-+ f(x 0) as k-+ oo. But also z;k-+ z 0 as k-+ oo, so that
z 0 = f(x 0 ), that is, z 0 Ef[D]. I
Corollary Let D be a compact set in tnn and let f:D-+ tn be continuous

on D. Then f(x) assumes both a maximum and a minimum value on D.
Proof Since f[D] is bounded by Theorem 2, SUPxeD f(x) = M is finite.

Ltt {f(x1)} be a sequence in f(D] such that f(x 1) -+ Mas j---+ oo. Since f[D}
is closed, we see that M Ef[D]. This shows that M = maxxevf(x). The proof
for the minimum is similar. I
Let us recall the case of an ordinary, continuous function f : [a, b}-+ tn.
For locating points x 0 of maximum or minimum values ofj(x), there are three
190 CluJpter Five Limits and Continuity in n Dimensions
alternatives:
(I) x 0 is an end point a orb, or
(2) f' (x0) = 0, or
(3) f' (x 0) does not exist.
A similar situation prevails for functions of several variables; for simplicity we

consider only the case n = 2.
Theorem 3 Let f be a continuous real-valued function defined on a compact

set D c Bl2 If the maximum (or minimum) value offoccurs at a point (x0 , y 0) e.D,
then
(1) (x0 , y 0 ) is a boundary point of D, or
of (xo, Yo)
(2) ;- d f(x, Yo) 1
= d- = 0 and
uX X <~:=<~:o
Of
uy
(xo, Yo) = dd f(x
Y
0, y) I
v=vo
= 0, or
(3) one or both of the above derivatives does not exist.t
Proof Unravelling the logic, we see that what we must prove is: if (x 0 , y 0 ) is
an interior point of D (that is, N.(x0 , y 0 ) c D for some e > 0) and if the partial
derivatives
of of
ox (xo, Yo) and oy (x 0 , y 0)
exist, then these derivatives vanish. But this is fairly obvious; consider, for
example, the functionf(x, y 0) (y 0 being fixed). This function of xis defined for
lx - x 0 1< e and has a maximum at x = x 0 Hence its derivative, which is
pecisely of! ox, vanishes at (x0 , y 0 ). Similarly ofjoy vanishes at (x0 , y 0). (See
Figure 5.10.) I
t The expressions offox(xo, Yo) and offoy(xo, Yo) are called the partial derivatives off
with respect to x andy, respectively, at the point (xo, y 0). Note that off ox is calculated simply
by differentiatingf(x, y) with respect to x, treating y as a constant.
graph of z =f (x, y.,)
Figure 5.10
Example
Locate the maximum and minimum of the function f(x, y) = x2 +y on the

ellipse x 2 + 4y2 :;;: 4.
Solution Note first that the inequality x 2 + 4y2 :;;: 4 determines a closed,
bounded (that is, compact) set, so thatfhas a maximum and minimum.
Checking for vanishing of partial derivatives, we have
of = 2x and of = 1.
ox oy
Since offoy never vanishes, and since offox and of!oy exist everywhere, the
only possibility is that the maximum and the minimum occur on the boundary
curve x 2 + 4y 2 = 4.
Substituting x 2 = 4 - 4y2 , we are left with the problem of finding the
maximum and minimum of the function
g(y) = 4- 4y2 + y (-1 :o;;y:;;: 1).
Solving this simple problem of ordinary calculus, we find that g(y) has a maxi-
mum at y = 1/8, with g(1/8) = 65/16, and a minimum at the end pointy = -1,
withg(-1) = -1.
Therefore,
fmax =J( ~: 3 D =~!'

fmin =J(O, -1) = -1.
We return briefly to theoretical considerations.
Theorem 4 Let D be a compact set in !Jn and let f: D -+ !Jm be continuous

on D. Then f is uniformly continuous on D.
Theorem 5 Let D be an arbitrary set in !Jn and let {/;} be a sequence of

functions from D into !Jm. Then the sequence {/;} converges uniformly to some
function f: D -+ !Jm if and only if the sequence {/;} satisfies the uniform Cauchy
condition on D, viz.
dn(f;,fk) = sup ll f 1(x) - fk(x) l! -+ 0 as j, k-+ oo.

zeD
Theorem 6 Let D and{/;} be as in Theorem 5. If each function/; is continuous

on D, and if/;-+ funiformly on D, thenfis also continuous on D.
The proofs of these three theorems (as well as the required definitions of
uniform continuity and so forth) are virtually identical to the corresponding
proofs in Chapter 3. You are asked to locate these proofs and carry out the
required modifications.
Theorems 5 and 6 will be used in Section 5.5 to construct some extremely
bizarre continuous functions.
Exercises
1. Locate the maximum and minimum values of

(a) f(x,y) = 3x- yon the square - 1::::;; x,y::::;; 1.
(b) f(x, y) = xy on the disk x2 + y 2 ::::;; 1.
(c) f(x, y) = xye-z-11 on the first quadrant x, y ;;::::: 0.
(d) f(x, y) = x2 + y 2 on the straight line y; + 2y = 1.
2. Show that a curve joining two points in !Jn is a compact set. (Literally
speaking, this is false; correct this statement and then prove it.)
3. Let D be a compact set and Ietx0 :D. Definef:D-+ f!ll byf(x) = ll x - x 0 I -

Show that min.,enf(x) exists and is positive.
The number p(x0 , D) = min.,en llx - x 0 ll is called the distance from x 0 to

D.
4. Let p(x0 , D) be as in Exercise 3. Show that
(a) p(x0 , D) ::;: ll x 0 - x 1 ll + p(x 1, D); and therefore,
(b) p(x0 , D) is a uniformly continuous function of x 0
5. Let f: D -+ f!Jlm, where D is a closed unbounded set in f!Jln, and f is contin-

uous on D. Assuming that f(x)-+ 0 as x-+ oo (x E D), prove that f is
uniformly continuous on D. (Hint: Given c.> 0, first pick M such that
11/(x) ll < s/2 for ll x ll ;;::.: M, xED.)
6. Let D be a closed, unbounded set in f!Jln, and let f be a continuous, non-
negative real-valued function on D. Assuming that f(x)-+ 0 as ll x ll -+ oo
(x E D) and thatf(x0) > 0 for some point x 0 E D, show thatfassumes a
maximum on D. Doesfnecessarily assume a minimum on D?
5.5 Two Famous Examples
In this section we will construct

(1) a continuous function which is nowhere differentiable;
(2) an "area-filling" curve, that is, a continuous function f: [0, 1]-+ f!/l 2 ,
whose image consists of an entire (square) area.
These examples indicate that the notion of continuity is more mysterious than
might be expected. The twentieth century has seen tremendous development in
the understanding of continuity; a special branch of mathematics called
topology is devoted to its study.t
The original example of a continuous, nowhere-differentiable function was
devised by Weierstrass. We present instead an example invented by the con-
temporary mathematician B. L. Van der Waerden. Let h(x) be defined as
follows (see Figure 5.11):
ifO::;: X::;: f,
h(x) = {x
1 - X ift ::;: X ::;: 1' XJ:: i] c{)
and extend h(x) periodically to the whole real line, so that h(x) == h(x + 1).
t A good introductory text is B. Mendelsohn, Introduction to Topology, Allyn and Bacon

(1962).
194 Clwpter Five Limits and Continuity in n Dimensions
- /l-t I /
' ~ O tt- -,__() 1 ":~, 2 ~r:'-<; _

~ : ;__ .J )'- 1 I l l),,fl [!,,} f ,l {j v,;t1 [;\.1 [-,,1_.
Figure 5.11 The function h.
Theorem (Vander Waerden)
00
f(x) =I w-kh(10kx). (5.16)
k=O
Thenfis continuous for all x, but f'(x) exists for no x.
Proof It is easy to show that f is continuous: each function w-kh(lOkx) is

continuous, and the series (5.16) converges uniformly.
The proof that f' (x) does not exist is more complicated, although the idea
is simple enough: the function h(lOkx) is nondifferentiable at each point x =
n/2 lQ-k (nan integer), and the set of all such points is dense in f!l.
Let a = a0 a1a2a3 be an arbitrary given real number. Define a sequence
{x.,} as follows:
where
a., + 1 if a .. i= 4 or 9,
b = {a.. - 1 if a..
n = 4 or 9.
Then x .. - a= to-n, so that x .. -. a as n-. oo.
We are going to prove that
f(x ..) - f(a) is an {odd} integer if n is {odd} (5.17)

x .. - a even even
This implies immediately that lim.,_.,(f(x) - (f(a))/(x- a) does not exist,
which proves the theorem. To help in understanding the following proof, it is re-
commended that you work through Exercises 1 and 2 below before proceeding.
It follows from the definitions of x .. and h that
lOk-n ifO:::;;; k < n,

h(lOkx..) - h(lOka) =
{0 if k :2: n,
where the + or - sign depends on the values of k and n. Hence,
10-k( l()k'-n)
10
_k h(10kxn)- h(10ka) = o-n
1
= 1 if 0 :::;;: k < n,
x -a {
n 0 if k ~ n.
Therefore, finally,
f(xn) -/(a) = i 10_k h(10kxn) - h(1Qka)

Xn - a k=O Xn - a
n-1
= 1(1),
k=O
and this value is an even or odd integer, when n is even or odd, respectively. I
Weierstrass' original example was a trigonometric series,
<Xl
/(x) = 1 an COS bn1TX (Ia I < 1).

n=O
This seems even more audacious than Van der Waerden's example, because
each of the summands an cos bn 1TX is infinitely differentiable, but still the sum
is nowhere differentiable. Many important developments in analysis have arisen
from the study of trigonometric series.t
We come now to the "area-filling curve" of Peano. This example, like the
previous one, is constructed as the uniform limit of a sequence of continuous
functions, in this case curvesfn:[O, 1]-+ D, where Dis the unit square 0 :::;;: x,
y :::;;: Lt [ Cl 1 :}'L
The c~e h is defined as shown in Figure 5.12; it joins the point (0, 0)
to (1, 1). Each of the 9 segments ofh is taken as a linear function on the corre-
sponding interval
k-1
[ ~~
k]
',9- ' k =
/
1, 2, ... )/~
)
This describes h uniquely.
The curve / 2 is constructed by replacing each segment of h by a curve ex-
actly likeh itself, but scaled down suitably; see Figure 5.13. Thusf2 has 92 = 81
t See R. L. Jeffery, Trigonometric Series, a Survey, University of Toronto Press (1956).

+In the following discussion of Peano's curve, most of the mathematical details are left
to the reader to verify.
Figure 5.12 The curve ft. k~ \ q ')- j S
L
Figure 5.13 The curve f 2 Y10 ~ "' r,r ~ S
J.
o 1 h 1 c 1 ,e 1 b 1 { 1 0 1 el;
linear segments, and passes through the same 101'corni_p_s>~ts of h with the
same values t = kf~ of the parameter tat each corner point. It follows that
3
the number -./lj3 being the diameter of a squar~ of side 1/3.

The sequence fin} of curves is obtained by repetition of the above process;
thus / 3 consists of 93 linear segments, and so on. We have
1
"" rq, 'I qt1 !],.1

11
[li'VN/\ ,
rQ-j ),
i,//J (i
!/11 b
'
1
r J
,
f f. ,f ;,t
From this it follows that {fn} say /ties the uniform Cauchy condition, and
therefore fn ~ f uniformly as n oo, where f is a continuous curve in D.
We propose to show that the im geoff is the entire square D.
Let x E D. If xis the corner, oint of one of the squares of side 3-k obtained
in the construction of the curves {fn}, then x lies on the graphs of all the curves
fk ,fk+1, . .. and we have
for some t 0 E [0, 1]. Hence,f(t0) = x, so that x lies on the graph off
If x is not the corner point of any square obtained in the construction,
then for each n, x will lie in such a square Qn of side 3-n. We can suppose that
Qn+1 c Qn for all n, and we have
co
nQn = {x}. (5.18)
n=l
The curvefn passes through the square Qn as a diagonal; letfn(tn) = Xn denote

the center of this diagonal. Then limn~ao tn = t exists, and f(t) E Qn for all n.
From (5.18) we conclude that f(t) = x, so that x lies on the graph off This
completes the proof that the curve f "fills" the whole square D.
Since Peano's area-filling curve may leave you with the impression that
continuity is a worthless concept after all, we quote the famous Jordan-curve
theorem, named after C. Jordan (1832-1922). This theorem, which has an
interesting history offallacious attempts at proof, states that ifjis a closed plane
curve which is simple, then f cannot be area-filling, but instead cuts the plane
into two pieces, an "inside" piece and an "outside" piece (Figure 5.14). (The
curve f is said to be simple if t1 # t 2 implies that f(t 1) # f(t 2), that is, if the
II I
Outside
I I
i
Outside Inside
Figure 5.14 A Jordan curve.

curve does not intersect itself.) Peano's curve (which could easily be made
closed) is very far from being a simple curve: it intersects itself infinitely often.
\!\~'( ,c_ ;<: -~\ cA "'q;\e( tAi11) ~VIM P uf0e r [ ov- r> "+ )
Exercise;~
1. Using the notation of the Theorem of Van der Waerden and taking a = 0,
_V\
(a) find Xn(n = 1, 2, 3, . . .); + \U
(b) find h(lOkxn) - h(lOka);
0
f (x) -f(a)
(c) show that n = n.
Xn- a
2. Let a = 0.814. Proceed as in Exercise 1.

3. Let fn be the nth curve constructed in obtaining Peano's curve. What is
the length of fn?
4;. Show how to- construct a space-filling curve whose image is the cube
0 ~x,y,z:::;;; 1.
5. Show how to construct a triangle-filling curve, on the basis of the curve

g 1 of Figure 5.15.
Figure 5.15 The curve g1

Appendix
/, (o) h { / _,,)
' ') "- I -;;..) I"
(; ) - (-
I;), ) K '? "" ""-) [ I, D\C- "' J ;: _ l Of<- V\ .,. 1 i,., :::_ 0
<:.-"'
[0 , .r (,P
;;;; o-J ;' ' '- ,_1
' '!...
Vi- I
)(- !.,
-1"-th
(c) ~
o-1
D ' -:J - 2() (t) -- 1_. I .1.. .. ~ r, l
<- ..., N/ , 't.(l<'
:: V\( ~
'
[ Logic
Although mathematics is usually thought to be a completely logical subject,

many students of mathematics are quite uncertain about the basic principles of
logic. As used in mathematics, logic has two parts: the "statement calculus,"
which deals with logical connectives, such as "and," "or," "not," and
"implies"; and the "quantifier calculus," whi~h deals with the quantifying
phrases "for some" ("there exists") and "for all."
The proper use of quantifiers is especially important in the study of mathe-
matical analysis. The definition of limit, for example, uses three quantifying
phrases:" For every e > 0, there exists b > 0, such that for every x, ... ."
Sentences of this degree of logical complexity are seldom if ever encountered
in a nonmathematical setting. The purpose of this Appendix is to introduce
you to the most frequently used rules of statement and quantifier logic.
1.1 Logical Connectives
The study of logic is concerned with the truth or falseness of statements.

Fer example,
is a true statement (as could be proved from the axioms and definitions of
arithmetic). The statement
x 2 < 1,
201
202 Appendix One Logic
however, is neither true nor false, since it contains an unspecified variable, x.

In this section we consider only statements without variabl~s.
Let A be a given statement. Then we can form the negation of A, which we
denote by
,_,A= notA.
If A is true, then ,_,A is false, and moreover, if A is false, then ,_,A is true.
The operation of negation can be described by the following simple
"truth table."
ill F
T
Given statements A and B, we can form the conjunction
A &B = A and B.
Whether the statement A & B is true or false depends on the truth of falsity of
A and B. We define A & B to be a true statement if and only if both A and B are
true. The truth table for A & B is, therefore, the following.t
A B A&B
T T T
F T F
T F F
F F F
Another common logical connective is the disjunction
A vB = A or B.
In mathematical logic, this is always the inclusive disjunction, so that A v B
t The arrangement of the truth columns for A and B in this table is standard and is
used in all truth tables. It can easily be expanded to construct truth tables with several state-
ments A, B, C, .
is true whenever either A is true orB is true, or both: t
A B AVB
T T T
F T T
T F T
F F F
Finally we have logical implication:
A:::>B - A implies B if A then B.
The truth table for implication is
A B A==> B -
T T T
F T T
T F F
F F T
Some readers may disagree with this table at fir~t, especially with the last row,
which says that "false implies false is true." Thus the statement
3+1=7:::>6-1=8
is a true statement! (In fact we can prove it! Suppose 3 + 1 = 7. Then
6 + 1 = 3 + 3 + 1 = 3 + 7 = 10.
Therefore 6 - 1 = 6 + 1 - 2 = 10 - 2 = 8.)
A statement P constructed from various substatements A, B, C, ... ,
and which is true no matter whether A, B, C, .. . are true or false, is called a
tautology.
t Everyday language uses both inclusive and exclusive disjunction, so that ambiguity can
arise. To avoid ambiguity, some authors are reduced to using the phrase "and/or" to indicate
inclusive disjunctions.
Example 1
[A & (A => B)] => B is a tautology.

(This says that if A is true and A=> B, then B is true.)
To verify this, we construct the following truth table, using the given truth
tables for & and=>.
A B A=>B A & (A =>B) [A & (A =>B)] => B
T T T T T
F T T F T
T F F F T
F F T F T
Since the final column contains only "T's," the given assertion is a tautology.t
Two statements P and Q, constructed from the same substatements A,
B, C, ... , are said to be logically equivalent if they have identical truth tables.
In this case we write P =
Q, or P <;::> Q.
Example 2
,.._,(A v B) = (,_.A) & (,.._,B). (In words, A. v B is false if and only if A

and Bare both false.) To verify this we again construct a truth table.
T T T F F F F
F T T F T F F
T F T F F T F
F F F T T T T
Since the columns for .-(A v B) and (.-A) & (.-B) are identical, we have
proved the equivalence of these statements.
t Note that this verification is completely routine. In fact, it is easy to program a computer
to verify tautologies.
The foregoing logical operations are basic ingredients of all mathematics

that every student of the subject should be familiar with. The following ex-
ercises cover many of the standard uses of logical connectives encountered in
mathematics.
Exercises
1. Verify the following equivalences by means of truth tables.

(a) A => B = (,._,B)=> (,._,A).
(Note: (,._,B)=> (,._,A) is called the contrapositive of A=> B. This
tautology is the basis of proof by contradiction.)
(b) A v B = (,._,A)=> B.
(c) ,._,(A & B)= (,._,A) v (,._,B).
(This should be compared with Example 2.)
(d) ,._,(A =>B) = A & (,._,B).
(e) A v B == B v A.
(f) A v (B & C) - (A v B) & (A v C).
(Note that the truth table for. 3 statements A, B, C must contain
23 = 8 rows.)
{g) A & (B v C) == (A & B) v (A & C).
(h) A ~(B => C) = (A & B)=> C.
(i) (A v B) => C = (A => C) v (B => C).
(j) A => (B & C) = (A => B) & (A => C).
2. Show that the following statements are tautologies.

(a) A=> A.
(b) ,._,(A & ,._,A) (Law of excluded middle).
(c) [(A =>B) & (B => C)]=> (A => C) (Law of syllogism).
(d) [A=> (B & ,._,B)]=> ,._,A (Reductio ad absurdum).
3. If P => Q is a tautology, we say that Pis logically stronger than Q. Which

(if any) of the following statements is logically stronger?
(a) (A & B)=> C; (A=> C) & (B =>C).
(b) A=> (B =>C); (A=> B)=> C.
4. Show that P = Q holds if and only if (P => Q) & (Q => P).
5. The statement B => A is called the converse of the statement A => B. Show
that a statement and its converse are not logically comparable.
6. Show that the logical connectives of implication and conjunction (for ex-
ample) could both be dispensed with in logic.
7. Give truth tables for each of the following assertions.

(a) either A or B but not both.
(b) A is true unless B is true.
(c) A is a necessary condition for B.
(d) A is a sufficient condition for B.
(e)_ A and !J are independent.
8. Express the assertions of Exercise 7 in terms of&, v, "' =>.
9. Given that 3 +1= 7, "prove" that 1 = 0, and hence that all numbers are
equal.
1.2 Quantifiers
LetS denote a given set; xES means that xis a member of S. Let A(x)
denote an assertion about x; for each choice of x, the assertion A(x) is either
true or false.
We can consider the statement
I. For every xES, A(x) is true, which we notate as
I. V XES, A(x).
The phrase "for all," or "for every," (V) is called the universal quantifier. The
statement V xES, A(x) is said to be true provided, naturally, that A(x) is true
for every choice of x inS. If Sis a finite set, say
then V xES, A(x) means the same as
If Sis an infinite set, there is no such simple interpretation.

We are also interested in the statement
II. For some xES, A(x) is true, which we abbreviate as
II. 3:xES,A(x).
1.2 Quantifiers 207
The phrase "for some," or "there exists," (3:) is called the existential quantifier.
The statement 3: xES, A(x) is true provided that A(x) is a true statement for
at least one x in S. If Sis finite,
then 3: xES, A(x) means the same as
The universal and existential quantifiers are thus seen to be extensions of

the logical connectives & and v, respectively, to deal with infinitely many
assertions, or assertions about infinitely many "things" x.
Example 1
Let FA denote the set of all real numbers, and !2 the set of all rational numbers.
We ask the reader to decide on the .basis of his experience in mathematics,
whether the following statements are true or false.
(a) Vx E FA, 2x > x Tor F?

(b) 3: X E fl, x 2
2= Tor F?
(c) VX E fl, x 2 E !2 Tor F?
(d) 3: X E FA, x5 - 3x2 +5= 0 TorF?
Note that (b) and (d) are not completely trivial!
The symbol "x" in the statements I, II is called a quantified variable. There

are two basic regulations regarding the use of variables in mathematics. The
first states that any variable which occurs must be quantified. For example,
x>2
makes no sense by itself, but only makes sense if the meaning of xis specified.
Another example:
x > 2=>x2 > 4;
this seems to be correct, because we are conditioned to interpret it as:
V x E FA, x > 2 => xs > 4. (1)

The latter statement makes sense and is true. Unwritten quantifiers of this kind
occur frequently in mathematical writing; they are acceptable provided the
context is absolutely clear.
The second rule is that a quantified variable is a "dummy" variable and can
be replaced (in all its occurrences) by any other variable symbol. For example,
Vt E fJl, t > 2 => t2 > 4
carries precisely the same meaning as Statement (1) above.

It is often useful to know how to negate a quantified statement. For this
we have the rules
(i) ,..._,(V XES, A(x)) = 3 xES, ,..._,A(x),
(ii) ,..._,(3 XES, A(x)) = V XES, ,..._,A(x).
Rule (i) is the rule of "counterexample." Rule (ii) can be derived from Rule (i).
Note that if Sis a finite set, rules (i) and (ii) are just known tautologies; for
example, supposeS= {xi> x 2}. Then
,..._,(V xES, A(x)) = ,..._,(A(x1) & A(x2 ))

= (,..._,A(xJ) v (,..._,A(xJ) = 3 xES, ,..._,A(x).
Other combinations of quantifiers and connectives will be found in the Exercises.
In mathematical analysis it is common to encounter statements involving
several quantifiers. For example, the definition of the limit of a function uses
three quantifiers: lim.,-+~~f(x) =Lis defined to mean
Y e > 0, 3 ~ > 0, Y x E fJl, (0 < lx- al < ~ => lf(x) - Ll < e). (2)
In handling such statements, it is essential to understand first that the order in

which quantifiers appear affects the meaning of the statement.
Example 2
Consider the statements

(a) VxefJl, 3yefJl, x+y=1,
(b) 3 y E fJl, V X E fJl, X+ y = l.
Statement (a) is true: given x E fJl, we can sety = 1- x, and then x y = l. +

Statement (b), however, is false: there does not exist ayE fJl such that x +
y = 1 holds for all x E fJl.
1.2 Quantifiers 209
To summarize, in any statement
VxES, 3:yET, A(x,y),
the choice of y is allowed to depend on the value of x, whereas in the statement
3:y E T, V XES, A(x,y),
the choice of y must be independent of x.

More generally, we can state the following rule: if an existential expression
3: x E T occurs in a statement, then the choice of x may depend on any vari-
able that occurs before x in the statement, but not on any variable that occurs
after x.
The negation of a statement with several quantifiers can be found by
repeated application of Rules (i) and (ii) above.
Example 3
Consider the (false) statement (b) of Example 2. The negation of this statement
is:
"-'(ct y E f!A, V X E f!A, X +y = 1) - V y E f!A, "-'(V X E fJI, X +y = 1)
= Vy E f!A, 3: X E f!A, X+ y =:f= 1.
(We could prove that this negated statement is true as follows: let y E &I. Take
x = - y; then x +
y = 0 =:f= 1.)
Exercises
1. Determine which of the following statements are true. (&I denotes the set of
real numbers, JV the set of all integers.)
(a) V X E f!A, 3:y E f!A,y 2 ~ x;

(b) 3: X E fJI, V y E fJI, y 2 ~ X;
(c) d y E f!A, V X E f!A,y ~ x; 2
(d) V X E f!A, cty EJV, X :s;;y;

(e) VxE&I,VyE&I,xy=yx;
(f) V e > 0, 3: <5 > 0, V x E &I, lxl < <5 => x 2 <e.
2. Consider the assertion "every real number is smaller than some integer."
Express this in formal notation, and decide whether it is true or false.
3. Let S be a given set. Decide (intuitively) which of the following assertions

are valid.
(a) [V XES, p(x)] ~ [3: XES, p(x)].

(b) [V xES, p(x)] & [x0 E S] ~ p(x0 ).
(c) {[V x E S, p(x)] & ,_,[x0 E S]}=> ,_,p(x0 ).
(d) [V XES, p(x) V q(x)] ~ [V xES, p(x)] V [V xES, q(x)].
(e) Same as (d), with v replaced by &.
(f) [V X E S,p(x)] V [V xES, q(x)] ~ V x E S,p(x) V q(x).
(g) [V XES, p(x)] & [3: XES, p(x) ~ q(x)] ~ [3: XES, q(x)].
4. Negate the following statements; decide whether each given statement or its
negation is true.
(a) V X E.Af, 3: y E.Af, y2 =X.
(b) V X E.Af, 3: y E.Af, Y = x2.

(c) VxEBI, 3: y E &/, xy = 1. '
(d) 3: X E &/, V y E &/, xy =0.
1
(e) V X E &/, V y E &/, xy =I= 0 ~ - + -1 = -1- .
x y x+y
(f) 3: C E &/, V X E &/, (x ~ C ~ 3: y E &1, l +y = x).
1.3 Proof
I
;:.
Mathematics is a deductive science. Its essence lies in the proofs of theorems.

This is not to underestimate the importance of intuition and insight in mathe-
matics. The final product of mathematical creation, however, must be a rigorous
logical theory.
What is a proof? This question frequently troubles beginning students of
serious mathematics. (For example, can you prove that 2 + 2 = 4? This fact
is certainly not an axiom or a definition, so it must be a theorem.) Let~.> con-
sider first an assertion of the form A ~ B, where A and B are given statements.
1.3 Proof 211
By a (formal)proojofthe assertion A=> B, we mean a finite list of statements
with the properties

(1) A 1 =A (the hypothesis) and An= B (the conclusion);
(2) Each statement A; is either an axiom or a logical consequence of the
previous statements A1 , A2 , , A;_1
Needless to say, "proofs" satisfying this definition are seldom encountered in

practice. This is because we have invented several ways of simplifying the writing
of proofs, such as (a) introducing definitions; (b) using known theorems with-
out reproving them; (c) omitting steps in the proof which the reader is expected
to fill in. We only understand the proof of a certain theorem when we are con-
vinced that we could (upon demand) write out a complete formal proof of it.
Consider next the case of a quantified statement of the form V xES, p(x).
To prove this assertion, we prove instead the assertion xES=> p(x), according
to the definition of proof just given.
Example
Theorem VX E (0, 1), X2 <X.
Proof: 1. Let x E (0, 1). (hypothesil.)

2. Then X > 0& X < 1. (definition)
3. Therefore, x x < x 1. (axiom)
4. But x x = x 2 , (definition)
5. And x 1 = x. (axiom)
6. Therefore x 2 < x. (substitution)
Finally, consider an existentially quantified statement 3: xES, p(x). To

prove this directly is to exhibit a member xES for which p(x) is true. For
example, the statement 3: x E &l, x 2 = 2 is proved in Chapter 1 by defining
x = limn-co Xn for a certain sequence {xn}.
We can also prove the existential assertion 3: xES, p(x) by the method of
contradiction. Thus we would show that the negation, namely, V xES,
"-'p(x), leads to a contradiction. According to the law of the excluded middle
(,_,""'A == A), this is a valid proof. (Some mathematicians do not feel satisfied
with such proofs of existence by contradiction, because proofs of this kind are
sometimes "nonconstructive," in the sense that no method of actually deter-

mining a value of x for which p(x) is true can be deduced from the proof. From
a practical point of view, this is certainly a valid criticism.t)
This completes our discussion of mathematical logic. It should be stated
that we have greatly oversimplified the subject.t Nevertheless, the principles
discussed here, if combined with one's logical intuition, should suffice for most
contemporary mathematics.
Exercises
Prove the following theorems. At each step indicate whether the statement is an
axiom, definition, known theorem, or a logical step. Identify all tautologies
used in the logical steps.
I. 2+2 = 4.
2. V n e.AI', n odd~ n2 odd.
3. V x E f!l, lxl 2 = x2 (Use the definition !xi = +x if x ;: :.: 0 and lxl = -x

if x < 0. Consider cases.)
4. ,_,(3: X E f!l, X 2 = -1).
].
1.4 Sets and Functions
The most basic notions in mathematics, at least by contemporary stand-

ards, are the concepts of a set and of membership in a set. For example, the entire
real number system can be built up by successive set formations, beginning only
with the single number zero. Further set formation leads to n-dimensional
spaces, functions, and so on, to all of calculus and analysis. We have no
intention of carrying out such a program here; rather we will simply discuss the
most elementary and useful properties of sets.
If X is a given set and p(x) an assertion which applies to any member x
t The definitive work on this topic is E. Bishop, Foundations of Constructive Analysis,

McGraw-Hill (1967).
~For further details seeR. R. Stoll, Set Theory and Logic, W. H. Freeman (1961); or
A. Tarski, Introduction to Logic, Oxford University Press (1941).
of X, we use the notation

{x EX lp(x)}
to denote the set of all members of X for which p(x) is true. For example,
{x E ~I x 2 < x} (read : the set of all x in ~such that x 2 < x) can easily be
seen to be equal to the interval
(0, 1) = {x E ~I 0 < x < 1}.
Let A, B, ... denote subsets of a given "universal" set X; in any given

situation the set X would be clearly defined. We can then perform the following
set operations:
1. I
Union: A U B = {x EX (x E A) v (x E B)}.
2. Intersection: An B = {x EX (x E A) & (x E B)}.I
3. Complement: A' = {x EX ,..._,(x E A)}.I
Note that these operations correspond to the basic logical connectives "or,"
"and," and "not." The connective "implies" also has a role in set theory, as
follows .
Definition 1 We say that A c B (A is contained in B) provided that
Vx E X, (X E A) => (x E B).
Furthermore, we define A = B to mean that A c B and B c A ; in other

words, A = B if and only if
V X E X, (x E A)<=> (x E B).
The following useful set identities are simply set-theoretic reflections of

various logical tautologies discussed in Section 1.1.
(i) (A')' = A.
(ii) A u A'= X,
A n = 4> (the empty set).
A'
(iii) A u
(B u C) = (A u B) u C,
A n (B n C) = (A n B) n C.
(iv) (A U B)'= A' n B',
(A n B)' = A' u B'.
(v) A u (B n C) = (A u B) n (A u C),
A n (B u C) = (A n B) u (A n C).
For example, here is how (iv) is proved:

x E (A U B)' == ,...._,(X E A U B)
- ,...._,((x E A) V (x E B))
= (,...._,x E A) & (,...._,x E B) (tautology)
=xEA' &xEB'
==xEA' nB'.
The remaining identities are proved in a similar way.

Frequently we wish to operate not with two sets, but with an arbitrary,
I
perhaps infinite, family of sets. Let {A..t A E L} denote an arbitrary family of
subsets of X; the set Lis called the "index set." Then we can define:
1. Union: U..tEL A..t 13 A E L, x E A,t}.
= {x E X
2. Intersection: n..tEL A,t = {x EX I v A E L, X E A,t}.
In other words, x belongs to the union of the family of sets A..t if and only if x
belongs to at least one set A..t; and x belongs to the intersection of the family of
sets A..t if and only if x belong to every set A..t.
Using the quantifier calculus of Section 1.2, we can obtain analogues of the
identities (iii), (iv), and (v) for the case of arbitrary families of sets; see Exercise
3.
Finally let us give the modern set-theoretic definition of "function."
Definition 2 Let A and B be given (nonempty) sets. A set f of ordered pairs

(a, b) with a E A and b E B, is called a functio.'1 from A to B, provided that for
every a E A there exists a unique bE B such that (a, b) Ef In case (a, b) Ej,
we write f(a) = b.
The set A is called the domain off, and B is called the range off
Iff is a function from A to B, we write
f:A---+-B.
Example
Let A= B = !?l an_d consider the setf of all ordered pairs (x, x 2) for x E !J.
This set satisfies the definition of a function, and we writef(x) = x 2
Definition 2, which may seem mysterious on first sight, actually agrees

with most people's understanding of "function." (Most calculus books, how-
ever, refer to the set of pairs (a, b) E f as the graph of the function.)
Exercises
1. Prove the set identities (i), (ii), (iii), and (v), identifying the tautologies
used.
2. LetA;.= {xEfJI~~ ~x ~A.}.If

I
L 1 = {A. E fJt A > 0} and L 2 = {A. E fJt 12 < A< 5},
find
n A;.
J.eLi
and U A;.
).eLi
(i = 1, 2).
3. Prove the following statements.
(b) ( U
J.eL
A;.) 11 B = U (A;.
J.eL
11 B).
4. UA;. =?
).e.
5. Letfbe the set of all pairs (x, y) with x E EJt, and with
1 if the digit 7 appears in the decimal

y = expansion of x,
{
0 otherwise.
Calculate /(t),JCf),f(/2).
6. Let f/ denote the set of all sets. Consider the set
T = {x E f/ I,_,(x Ex)}.
Show that TE T~,_,(TE T).

This is called Russell's paradox, after Bertrand Russell (1872-1969).
The moral is, roughly speaking, that the "set" of all sets isn't really a set
itself because it's too "inconceivable." Set theory is today a major area of
mathematical research, and there have been recent revolutionary develop-
ments.
II Mathematical Induction
Jl.l Mathematical Induction
Consider the following simple problem.
Example 1
Find the value of

1+4 + 9 + 16 + ... + 100.
(What are the missing terms, represented by 3 dots?) Now find the value of
1+4 + 9 + 16 + ... + 10,000.

No sane person would try to solve this by arithmetic-he'd look for a more
"mathematical" method. If he looks in the right book (for example, this one)
he'll find the following formula:
J2 + 2 + + + n = -fin(n + 1)(2n + 1).

2
32
2
(11.1)
The values of the sums given above are 385 and 338,350, respectively.
The question now is, how can Formula (11.1) be derived? The truth is that
it's tricky to discover such formulas, although guessing sometimes works.
Let's suppose we obtain (11.1) by guessing. Next question: is the guess correct?
216
11.1 Mathematical Induction 217
First we can check a few special cases (we may soon discover if (11.1) is wrong!).
n 2 3 5 IO etc.
I2 + 22 + 32 + ... + n2 5 14 55 385 ?
tn(n + I)(2n + I) 5 14 55 385 ?
Is this table a proof of Formula (II. I)? No. (Remember the "physicists' proof"
that every odd integer is a prime: "3 is a prime, 5 is a prime, 7 is also, 9 is an
experimental error, but II is prime, 13 is prime, and so on.")
Here's how to prove (11.1); note that we are trying to prove that (11.1)
holds for every positive integer n. Suppose not. Then there exists a positive
integer N 0 such that (11.1) is false for n = N 0 Let N denote the first (smallest)
positive inte.ger for which (II.I) is false. Then:
(a) N > I, because (11.1) is surely true if N = 1.
l(b) Formula (11.1) is true for n = N- 1.

(c) It's false for n = N.
Accordingly, by (b).
.
12 + 22 + 32 + + (N- I) 2 = t(N- I)N(2(N- I)+ I).
Adding N 2 to both sides, we have
12 + 22 + 32 + + (N- I) 2 + N2 = t(N- I)N(2N- 1) + N 2

= iN[(N- I)(2N- I) + 6N]
= tN[2N2 + 3N + I]
= tN(N + I)(2N + 1).
But this shows that (11.1) is true for n = N, in contradiction to (c). Therefore,
we have proved that Formula (11.1) must hold for every n.
The proof just given is an indirect proof by mathematical induction.
("Indirect" means by contradiction.) We could have given a direct proof, as
in Example 3 below.
Example 2
It is well known that the sum of the interior angles of any triangle is I80-see
Figure A.l.
218 Appendix Two Mathematical Induction
------?i--~-y
_# _______ _
a {3
----- ' -----
a+ {3 + y = 180
Figure A.l
(a) (b)
Figure A.2
Question: What is the sum of the interior angles of a convex polygon with n
sides? A good student should be able to figure this out, so read no further until
you've tried.
Now let us try to guess the answer. Taken = 4 first. By Figure A.2(a), the
sum of the interior angles is obviously 2 x 1&0 (why?). But now take n = 5
(Figure A.2(b)). The interior angles obviously total 3 x 180. We can guess that
generally :
The sum of the interior angles of an n-sided convex polygon (n ~ 3)

equals (n - 2) x 180.
But this is more than a guess-we can easily see why it is true for every n.
Namely, every time we increase n to n + 1 we add another triangle, so that the
interior angles increase by 180. Since the formula (n - 2) x 180
(a) is correct for n = 3, and

(b) is correct for any value n + 1 whenever it is true for n,
we must conclude that this formula is correct for all values of n, n ~ 3.
The foregoing argument is a simple example of a (direct) proof by mathe-
matical induction. In general, the method of proof by mathematical induction
is based on the following axiom.
Axiom of Mathematical Induction Let Pn represent a formula, or other

statement, concerning the positive integer n. Then P n is true (or all positive
integers n provided that
I (i) P 1 is true, and

(ii) whenever Pk is true (for some k :::2: I), then so is Plc+ 1 .t
We call this an axiom because it cannot be proved on the basis of simpler
axioms. (For an indirect formulation of the axiom of mathematical induction,
see Exercises I5 and I6.) Most students will probably agree, at least after a little
study, that the axiom is intuitively "obvious"-it really describes a property of
the system {I, 2, 3, ... } of positive integers which we automatically believe.
Here is how the axiom of mathematical induction is used in practice.
Example 3
I + 2 + 3 + + n = !n(n + I). (11.2)
Let us prove this famous formula by induction :t

(i) P 1 is true, because P 1 says 1 = t I (2).
(ii) if Pk is true, then
I + 2 + 3 + + k = !k(k + I).
Adding k + I to both sides, we have

I + 2 + 3 + + k + (k + I)= !k(k + I)+ (k + I)
= Hk + I)(k + 2).
This proves that Plc+ 1 is true (why?). Therefore by the axiom, P n is true for all n.
Notice how similar this direct proof is to the indirect proof we gave in Example
I.
Let us repeat: in order to prove the statement "Pn is true for all integers
n :::2: 1" (that is, "V n 2 I, P n") by mathematical induction, we must do the
following:
(i) check that P 1 is true,
(ii) prove that for every integer k :;;::: I, if Pk is true then so is Pk+l
t In precise logical terms, Condition (ii) says: for all k ;::::: 1, Pk implies Pk+t
t "Induction" means the same as "mathematical induction."
In the exercises below, we use the summation notation. The expression

2;'=n ak de~otes the sum of ak from k = n to- k = m;
m
.2 ak = an + an+l + ... + am. (11.3)
k=n
Exercises
1. Write out in full (every term of) the following sums.

4 1 8
(a) .2 -:'
i=l}
(b) ,22\
k=5
3 PTT 10
(c) _2sin-,
2
(d) .2 1.
2>=3 m=O
2. Evaluate each sum in Exercise 1.
3. Show that
Therefore, the variable of summation is a "dummy" variable. It's like ;:c in

f!f(x) dx.
4. Write out an "indirect" proof of Formula (11.2).
5. Note that
1 = 1,
= -(1 + 2),
1- 4
1 - 4 + 9 = 1 + 2 + 3,
1 - 4 + 9- 16 = -(1 + 2 + 3 + 4).
Guess the general law, and prove it by induction.
6. The following summation formulas are to be proved by induction. You

may wish to write out some of the sums longhand-for example,
n
2 (2k - 1) = 1+3 + 5 + + (2n - 1).
k=l
n ,
(a) ,2(2k- 1) = n 2
k=l
n 1 - an+l
(c) 2ak = (a =j::. 1).
k=O 1 -a
7. Show that for a =I= 1
a
!n kak = [nan+l - (n + 1)a + 1].
11
k=l (1 - a) 2
(Can you do this in two ways, one not using induction?)
8. Prove that if 0 < x < 1, then 0 < xn < 1 for every positive integer n.
9. Prove that if e > 0, then
(1 + e)" ;::::: 1 + ne (n = 1, 2, 3, .. .).
10. Show that if n straight lines are drawn in the plane, then the total number
of intersections cannot exceed !n(n - 1).
11. Using the product rule of calculus, D(uv) = uD(v) + vD(u), show by
induction that
D(u") = nun-1 D(u) (n = 1, 2, 3, . .. ).
12. For which integers n is 2n > n2 ? Prove your answer by induction.
13. Show that every positive integer is interesting.
14. Inductive proofs don't always have to begin with n = I; often we wish to
prove a statement: P .. for all n ;::::: N 0 (See Example 2 and Exercise 12.)
State a more general form of the axiom of mathematical induction to cover
such cases. Can you prove that the general form is a consequence of the
original form?
15. Consider the following axiom:

Axiom X Every nonempty subsetS of the set of all positive integers contains
a least element.
Prove that Axiom X implies the axiom of mathematical induction. (Hint :

Let Pn be a given statement and suppose P n satisfies conditions (i) and (ii) of
the axiom of mathematical induction. Suppose that P n is not true for all
integers n ;::::: 1. LetS denote the set of integers for which P n is false. Obtain
a contradiction.) Note that Axiom X was used implicitly in Example 1 of
this section .
*16. Show that the axiom of mathematical induction implies Axiom X, so that
these two axioms are logically equivalent.
*17. Determine, as a function of n, the maximum possible number of subregions

of the plane which can be formed by drawing n straight lines (extending to
infinity in both directions) in the plane. (It's not 2n.)
*18. Show that the sum of the digits in any multiple of9 is divisible by 9.
19. n straight lines (extending t<;> infinity in both directions) are drawn in the
plane. The resulting configuration is to be colored like a map-no two
"countries" with a common border may have the same color (but two
countries which meet at a single point can have the same color). Prove that
only two colors are needed.
II.2 Mathematical Induction (Continued)

It is often convenient to use mathematkal induction in definitions.
Example
Let a sequence {xn} be defined by the following "recursion formula" (see

Chapter 1).
Xn+l = 2xn + 1 (n ;;::: 1).
It is easy to calculate a few values of xn and then to guess and prove the general
formula:
x3 = 7,
and we can guess that in general xn = 2n - 1. We leave it to the reader to prove

(or disprove) this guess by induction.
A sequence defined by means of a recursion formula is sometimes said to

be "defined inductively." Some further examples are given in the exercises.
Next we consider another form of the axiom of mathematical induction
which is sometimes useful.
Axiom qf Mathematical Induction Strong Farm A statement P n is true

for all integers n provided that
(i) P1 is true, and
(ii) whenever P1 is true for all j < k, then so is Pk true (for k = 2, 3,
4, .. .).
1/.2 Mathe11Ultica/ Induction (Continued) 223
This is called the "strong form," because it is generally easier to verify

(ii) in this form than (ii) of the earlier axiom. Here is an example; this theorem
would be very hard to prove using the first axiom directly.
Theorem Every integer 2 2 can be written as a product of primes.
Proof (Remember that, by definition, n is a prime if and only if the only

integers which divide n are 1 and n.) Consider the statement Pn: the integer n is a
product of primes. Obviously P 2 is true. To prove that (ii) of the strong form of
the axiom of mathematical induction holds, let k be a given integer > 2 and
suppose that every integer j < k is a product of primes. There are two cases:
(a) if k itself is a prime, then Pk is true.
(b) if k is not a prime, then k = I m, where I and m are < k. Hence, by
assumption, both I and m are products of primes. Therefore so is k a
product of primes.
This proves (ii) and therefore we conclude that P n is true for all n. I
Exercises
1. Define Xn (n = 1, 2, 3, ... ) by
1
x1 = 1; Xn+l = Xn - n(n + 1) .
Determine xn explicitly.
2. Define {xn} as follows:
1
Xn+l = 1
+ Xn (n 2 1).
Check that the first few terms of this sequence are {1, t, f, t, f, ... }. Now
show by induction that
where {Zn} is the Fibonacci sequence, which is defined inductively as

follows:
3. Define the symbol !;=1 ak by induction. (The sequence {ak} is assumed given
in advance.)
4. Define n! by induction.
5. Define x1 = 2 and xn+l = 2"'" (n :;;::: 1). Find x 2 , x 3 , x 4 Show that x 6

cannot be calculated by hand in a lifetime.
6. Prove the strong form of the axiom of mathematical induction from the
original form (Section ILl). (Hint: Let Pn satisfy the hypotheses of the
strong form, and let Qn denote the statement "Pk is true for 1 :::;: k :::;: n."
Show that Qn is true for all n.)
11.3 The Binomial Theorem
Wanted: a formula for (a + b )n, valid for all positive integers n. Examples:
(a + b) = a + b,
1
(a+ b) = a + 2ab + b
2 2 2
,
(a+ b) = a + 3a b + 3ab + b
3 3 2 2 3
,
(a + b) = a + 4a b + 6a b + 4ab + b
4 4 3 2 2 3 4

The general formula will obviously be of the form
n
(a+ bt = 2Ckan-kbk;
k=O
the problem is to determine ck.

Notation: (1) If n is a positive integer, n! denotes
n!=123n;
also by definition
0! = 1.
(This strange definition is to make the binomial theorem simple.)

(2) If 0 :::;: k :::;: n, both positive integers, then
(~)
n!
k!(n- k)!
11.3 The Binomial Theorem 225
These 'numbers are called "binomial coefficients" because (we will show)
ck = (~)
Lemma
(11.4)
The proof (simple algebra) is left to you-see Exercise 2.
The Binomial Theorem
(11.5)
In this formula it is assumed that a and b are real numbers, and n is a

positive integer. The reader should check the formula for a few small values
of n, such as n = 3 or 4.
Proof By inspection, the formula is valid for the case n = I. To complete the
proof we use mathematical induction. Thus (changing notation slightly from
Section II.l) let us suppose (11.5) holds for n. We must deduce that the formula
holds for the next case, n + 1. We have
(a + b)n+l = (a + b)(a + b)n

=(a+ b),i (n)an-kbk (by assumption)
k=O k
This is indeed the formula (11.5) for the case n + 1, and the proof is complete. I
Example
Find the coefficient of x 5 in the expansion of (2 - x) 8
Solution The term in question is
( 8) 2a(- x )s = _!!____ 8( -1 )sxs

5 5! 3!
= -448x5 ;
the coefficient is -448.
There are other proofs of the binomial theorem. It is a simple consequence

of Taylor's theorem from the calculus, for example. The coefficients (~) are
sometimes denoted by nck, and nck is equal to the number of combinations of
n objects taken k at a time. A proof of the binomial theorem may be built on
the study of permutations and combinations.
Pascal's triangle (below) has a simple relationship with the binomial
theorem, which we leave the reader to elucidate. There is a generalization of the
binomial theorem that holds if n is not an integer. The sum in (11.5) becomes an
infinite series in general, as any standard calculus text will show.
(n = 0)
1 (n = 1)
1 2 1 (n = 2)
Pascal's triangle
1 3 3 1 (n = 3)
4 6 4 1 (n = 4)
1 5 10 10 5 1 (n = 5)
Exercises
1. Verify the following formulas.
(a) (~) = (:) = 1. (b) (~) = (n ~ 1) = n.

n'
(c) __: = n(n- 1) (k + 1) if k $;: n.
k!
(d) ~ = (k- 1)!. (e) (~) = (n ~ k)

11.3 The Binomial Theorem 227
2. Prove the Formula (11.4).
3. Find
(a) the 4th term in the expansion of (IX + ef3)1,
(b) the constant term in the expansion of ( x 2 - ~s.
(c) the coefficient of x 4 in the expansion of (x2 + y)
7
4. Estimate (1.02) 5 to 4 decimals.
5. Prove that
*6. Prove the following formula, called Leibniz' s rule.
Dn(uv) = i (n)k DkuDn-kv.

.k=O
(Dn stands for the nth derivative, that is, Dnw = ~:~ )
*7. State and prove the trinomial theorem.
Solutions to Selected Exercises
Solutions are given only for exercises that have explicit numerical or logical
answers, and not for exercises requiring proofs or estimates.
Chapter 1
Section 1.1
1. (a) 2, 4, 8, 16, .. . , (b) -1, +1, -1, ... ,

(d) -1, -2, -3, -2, -1, 0, +1, ....
2. (d) Xn = n2 + 1, (f) Xn = n! (n factorial).
3. (a) Xn = 2n- 1, (b) Xn = t{(-1)n+l + 1},

(d) Xn = n, (e) Xn = (-l)n+l.
Section 1.2
1. Xn+l = (x; + a)f(2xn).

4. The sequence {xn} is {1, 2, 1, 2, ... }, which diverges.
6. Xn+l = (2x! + 2)/(3x~).
229
230 Solutions to Selected Exercises
Section 1.4
1. (a) I, (b) I/4, (d) diverges, (e) 0.
2. (a) I, (b) I, (c) 0, (d) 0, (e) 0.
Section 1.7
1. (a) 2, 4, 6, (b) 2, 3, 4, 6, (c) 4, 6, (d) I, 5, 7, (e) 2, 4, 6.
I
IO. No. Example: xn = O,yn = -.
n
Section 1.10
3. +oo.
Section 1.12
1. (a) and (b) diverge; the rest converge.
Chapter 2
Section 2.1
3. (a) 2/3, (b) 2, (c) 0, (d) 0.
Se~tion 2.3
1. (a) - oo, (b) + oo, (c) -I/2, (d) + oo, (e) 0, (f) 0,
(g) 1Tj2, (h) -00.
2. (a) I, (b) 0, (c) I/2, (d) I/2.

Section 2.4
1. (a) Infinite discontinuities at x = 1. (b) Infinite discontinuity at

x = -1; removable discontinuity at x = + 1, with limit 1. (c) Jump
discontinuity at x = 1, with J 1 {f) = 2. (d) Infinite discontinuities at
x = mr, n = 0, 1, 2, . . . . (e) Removable discontinuity at x = 0,
with limit 0. (f) Same as (e). (g) Jump discontinuity (jump 1) at
every point x = n/10, n = 0, 1, 2, . . . . (h) Infinite discontinuity
at x = 0. (i) Removable discontinuity at x = 0, with limit 0; infinite
discontinuities at x =:= 1. U) Infinite oscillating discontinuity at
x=O.
3. Removable.
Section 2.5
2. (a) x = 0, (b) X= 0, (c) x = mr, n = 0, 1, 2, ... ,
(d) x = 0 and x = 2.
4. (a) -1, +1, (b) +1, -1, (c) +1, -1,
(d) 0, -2at x = 0, +2, 0 at x = 2.
Chapter 3
Section 3.2
1. (a) inf A = min A = 0, sup A = + oo, max A doesn't exist,
(b) inf A = min A = 7TT/22, sup A = max A = 10,
(c) inf A = 0, sup A = 1, neither min A nor max A exists,
(d) inf A= 0, sup A= max A= 1,
(e) inf A = 0, sup A = max A = 0.888 = 8/9.

3. (a) 1 (not max), (b) 0 (not min), (c) 2 = min,

(d) + 1 (not max), -1 (not min), (e) 1 (not max), (f) 0 =min,
(g) min = sin x 0 fx0 , where x 0 is the first positive solution of the
transcendental equation tan x = x (x 0 R:J 4.493).
inf A = min A = 1, sup A = max A = ~

1
6. [3
7. 3 (in both cases).
Section 3.3
I. (a) [-1, +1], (b) The empty set, (c) &I' (d) {0}.
Chapter 4
Section 4.1
l. ~ = af4; yes.
Section 4.2
1. U 11 (f)= (n +
1)(2n + 1)/6n2 ; L 11 (f) = (n- 1)(2n- 1)/6n2 ;
limits are both 1/3.
2. 7T = {0, 1, 2, ... , 10}.
Section 4.3
l. (a) 2e-4x",
3. G(x) = x if 0 ~ x ~ 1, and G(x) = 2x - 1 if 1 ~ x ~ 2. Notice that

G'(1) does not exist.
Section 4.4
1. p > -1.
2. p < -1.
5. (a) Diverges, (b) Converges, (c) Converges, (d) Converges,

(e) Converges.
6. (a) '"' (b) 1/2.
Section 4.5
1. (a) fn----+ 0 (pointwise); the convergence is uniform on / 2 only,

(b) fn(x) -1 if x =1- 0, andfn(O)----+ 0; the convergence is not uniform on
any of the intervals,
(c) fn(x)----+ 0 if x =1- 0, and fn(O) - 1; the convergence is uniform on / 3
only.
2. (a) f(x) = 0, (b) [nf(n + 1)]n+l, (c) No, (d) Yes.
Section 4.6
1. (a) 2, (b) 1, (c) +oo, (d) 2, (e) 2/3.
2. (a) 1, (b) 1/2, (c) 1, (d) +oo, (e) 1/2, (f) 1/e,
(g) 1, (h) Not a power series (converges for lxl > 1).
Section 4.7
1. (a) k~( -l)kx2k+1 j(2k + 1)!.

(b) k~( -1)kx2k I (2k)!.
(c) J 0
x 2k/<2k)!.
co
3. 'La 1x 1, where a 2k = ( -l)k(sin x 0)/(2k)! and a 2k+l = ( -l)k(cos x 0)/(2k + 1)!
0
4. (a) xf(I - x) 2 , (b) x(x + 1)/(1 - x) 3
7. (b) 0.90433.
8. 0.737.
Chapter 5
Section 5.1
4. (a) Closed, (b) Closed, (c) Open, (d) Neither,

(e) Neither, (f) Closed.
Section 5.2
5. (a) ?A3 , (b) 0, (c) gtn, (d) {0},

(f) fir(O) n B.
Section 5.3
3. (a) A conical helix, (b) An ellipse, (c) A straight line.
4. [This is a torus.]
5. f(t) = (1 - t)x1 + tx2 , 0 ::0:: t ::;;; I.
Section 5.4
1. (a) /max =/(1 , -1) = 4; /min =J(-1, 1) = -4,
(b) Let oc. = -./ ((-./ 5 - 1)/2). Then Jmax = J( oc,-./ (I - oc.2)) = oc-./ (I - oc.2)
and f min = !( -oc, .J (1 - oc2)) = -oc.J (l - oc) 2 ,
(c) f min= j(x, 0) =f(O,y) = 0 (for arbitrary x,y ::2: 0); /max= j(IJ.J2,
I//2) = lf2e,
(d) /min= j(l/5, 2/5) = 1/5; there is no max.
Appendix I
Section 1
7. The last columns in the truth tables are:

(a) FTTF, (b) TTTF (same as ,.._,B =>-A),
(c) TFTT (same as B =>-A), (d) TTFT (same as A=>- B),
(e) This is not a logical assertion.
Section 2
1. (a) T, (b) T, (c) F, (d) T, (e) T, (f) T.
3. (a) F if Sis empty, otherwise T, (b) T, (c) F, (d) F,

(e) T, (f) T, (g) T.
Section 4
2. (a) U A;.= (0, oo), (c) U A;.= G, 5),

L1 L2
(d) n A;. = [t. 21.

L2
5. (a) f(lf) = 0, (b) /(t) = 1,

(c) j(/2) = 1, since .J2 = 1.414213562373 ....
Index
Absolute convergence, 102 Composite function, 78

Absolute value, 7 Conditional convergence, 150
Accumulation point, 98, 178 Conjunction, 202
Alternating series, 51 Continuity, uniform, 109
Analytic function, 157 Continuous differentiability, 87
Angle, 171 Continuous, right, 81
Archimedean order, 30 Continuous function, 72, 182
Area, 116 at end point, 103
Area-filling curve, 195 Contradiction, 205
Averages, 162 Contrapositive, 205
Convergence:
absolute, 102
conditional, 150
Ball, 171 of improper integral, 134
Binomial theorem, 225 ofsequence,9, 17,176
]. Bolzano-Weierstrass theorem, 99, 17 8
of series, 44
Bound: radius of, 147, 149
greatest lower, 93 uniform, 140, 141
least upper, 93 Cauchy's criterion, 144
lower, 93 of power series, 152
upper, 93 Converse, 205
Boundary of a set, 17 4 Countable set, 41
Boundary point, 174 Counterexample, 208
Bounded function, 65, 96 Curve, 185
Bounded sequence, 23
Darboux, G., 115

Cantor, G., 40, 41, 42 Darboux sum, 115
Cauchy criterion, 102, 144, 178, 192 Dedekind, J., 71
Cauchy sequence, 101, 178 Dense subset, 42, 180
Cauchy's rem<1inder, 162 Decimal expansion, 30
Change of variables formula, 132 Denumerable, see Enumerable set
Closed set, 173 Derivative, 83
Compact set, 188 of an integral, 129
Comparison test, 48, 49, 50 partial, 190
Completeness property, 29, 95 right-hand, 90
237
238 Index
Differentiable function, 83 Hadamard, J., 149

continuously, 85, 87 Harmonic series, 46
Dirichlet's function, 76, 123 Hyperharmonic series, 48
Discontinuity, 72
Disjunction, 202
Distance, 8
between functions, 140 Image:
between vectors; 170 inverse, 184
Divergence: of function, 181
of sequence, 20 Implication, 203
of series, 44 Improper integral, 134
Divergent to infinity, 37 Induction, mathematical, 219, 222
Domain of function, 181, 214 Inequality:
Dot product, 171 and limits, 24, 64
Schwarz, 171
triangle, 7, 140, 171
Infimum, 93
e, 33 Infinite series, 44
computation of, 36 Inner product, 171
Enumerable set, 41 Integrable function, 120
Equivalence, logical, 204 Integral, 116, 127
Equivalent sets, 41 improper, 134
Euclidean n-space, 169 Integration by parts, 131
Euler's formula, 158 Interivr point, 190
Exponential function, 163, 166 Intermediate value theorem, 103
generalized, 106
Interval, 18
Factorial n, 224 Inverse image, 184
Function(s), 103, 181, 214 Iteration, 3
analytic, 157
bounded, 96
composite, 78
Jordan, C., 197
continuous, 72, 182
continuously differentiable, 85, 87
differentiable, 83
distance between two, 140 Kronecker, L., 6
domain of, 181
exponential, 163, 166
integrable, 120 Least upper bound, 93
logarithmic, 165 Leibniz's rule, 227
-nowhere differentiable, 193 Length of a vector, 170
trigonometric, 166 Limit:
uniformly continuous, 109 at infinity, 67
Fundamental theorem of calculus, of function, 59, 181
130, 131 uniqueness, 62, 182
of sequence, 9, 17
uniqueness, 19, 177
of sum, product, etc., 22, 63
Gamma function, 135 one-sided, 66
Geometric progression, 26 pointwise, 138
Geometric series, 49 Limit superior, 147
Greatest lower bound, 93 Logarithm, base of natural, 33
Index 239
Logarithmic function, 165 Real number system, 6

Logical equivalence, 204 Recursion formula, 2
Repeating decimal, 31
Riemann, B., 115
Mathematical induction, 219, 222 Riemann integral, 116, 127
Maximum, 93, 105 Rieht-continuity, 81
of function, 189 Right-hand derivative, 132
Mean value theorem: Rolle's theorem, 111
for derivatives, 113 Russell, B., 215
for integrals, 160
Minimum, 93, 105
Schwarz's inequality, 171

Negation, 202 Sequence, 1
Neighborhood, 59, 172 Cauchy, 101, 178
Newton's method, 5 limit of, 9, 17
N onincreasing sequence, 28 uniqueness, 19, 177
Nowhere differentiable function, 193 nonincreasing, 28
Number: Series, 44
irrational, 5 alternating, 51
rational, 5 geometric, 49
real, 6 harmonic, 46
Number system, real, 6 hyperharmonic, 48
power, 147
Taylor, 158
Set, 212
Obvious, 11 3 Space, Euclidean, 169
Open set, 173 Sphere, 171
Order, smaller, 83 Sum:
of series, 44
partial, 44
Partial derivative, 190 Supremum, 93
Partition, 115 Surface, 186
Pascal's triangle, 226
Peano, G., 187
Peano's curve, 195
Point, 169 Tangent, 84
boundary, 174 Tautology, 203
interior, 190 Taylor series, 158
Pointwise limit, 138 Taylor's theorem, 157, 159
Power series, 147 Triangle inequality, 7, 140, 171
uniform convergence, 152 Trigonometric functions, 166
Truth table, 202
Quantifiers, logical, 206

Uniform continuity, 109
Uniform convergence, 140, 141
Radius of convergence, 147, 149 Cauchy's criterion, 144
Range of function, 214 of power series, 152
Rational number, 5 Uniqueness of limits, 19, 62, 177, 182
240 Index
Vander Waerden, B. L., 193 Weierstrass, K., 71

Variable:
logical, 202
quantified, 207
Vector, 169
dot (inner) product, 171
INDEX OF SPECIAL SYMBOLS
r (x), Gamma function, 135

inf, infimum, 93
ff, the positive intet;ers, 41
!2, the rational numbers, 7
Bl, the real numbers, 6
Bin, Euclidean n-space, 169
sup, supremum, 93
{ Xn I , a sequence, 1
I I, absolute value, 7
II II , length, 170
of/ox, partial derivative, 190
~. summation symbol, 220
n!, factorial n, 224
( k) = nCk, binomial coefficient, 226

&, and, 202
V, or, 202
,....,, not, 202
=>, implies, 202
<=>, =, logical equivalence, 204
V, for all, 206
3 , there exists, 206
E, set membership, 6, 213
C, is contained in, 213
U, union, 213, 214
n' intersection, 213, 214

Theoretical Calculus by Clark 1972 Lite

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Theoretical Calculus by Clark 1972 Lite

Uploaded by

Copyright:

Available Formats

The Theoretical Side of Calculus

The Theoretical Side of Calculus

University of British Columbia

Robert E. Krieger Publishing Company

Printed and Published by

All rights reserved. No reproduction in any form of this

Printed in the United States of America

Library of Congress Cataloging in Publication Data

Clark, Colin Whitcomb, 1931-

A standard feature of the present mathematics curriculum at many

This text introduces students to mathematical analysis at as elementary a

1.2 The Method of Iteration 3

1.3 Properties of the Real Number System 5

1.4 The Limit of a Sequence (Introduction) 9

1.5 The Limit of a Sequence (Numerical Examples) 13

1.6 The Limit of a Sequence 16

1. 7 Elementary Theory of Limits 21

1.8 The Completeness Property 27

1.9 The Base of Natural Logarithms 33

1.10 Growth Properties of Certain Sequences 37

1.11 Some Further Properties of the Real Number System 40

1.12 Infinite Series 44

2.2 The Limit of a Function 59

2.3 Other Types of Limits 66

Chapter3 Properties of Continuous Functions

3.2 Supremum and Infimum 92

3.3 The Bolzano-Weierstrass Property 98

3.4 Cauchy Sequences 101

3.5 Properties of Continuous Functions 103

3.6 Uniform Continuity 107

Chapter 4 Some Theorems of Calculus

4.1 The Mean Value Theorem 111

4.2 The Riemann Integral 115

4.4 Improper Integrals 133

4.5 Uniform Convergence 138

4.6 Power Series 147

4.7 Uniform Convergence of Power Series 152

4.8 Taylor's Theorem 159

4.9 On the Definition of the Exponential Function 163

ChapterS Limits and Continuity in n Dimensions

5.1 Open and Closed Sets in Bin 169

5.2 Sequences and Limits 176

5.3 Limits and Continuity 180

5.4 Properties of Continuous Functions 188

5.5 Two Famous Examples 193

1.1 Logical Connectives 201

1.2 Quantifiers 206

1.3 Proof 210

1.4 Sets and Functions 212

II.l Mathematical Induction 216

II.2 Mathematical Induction (continued) 222

11.3 The Binominal Theorem 224

Solutions to Selected Exercises 229

t P. L. Wantzel (1814-1848) proved thilt cos 20 and t'2are nonconstructible. C. L. F.

the Greeks' concept of number was destroyed completely, and mathematicians

that a trigonometric series could converge to a discontinuous function such as

dy = limf(x +h)- f(x).

Cauchy explained the meaning of limit in the following terms:

When the successive values attributed to a variable approach in-

The intuitive idea of an infinite sequence,

Xl, X2, Xs, ' Xn, ' (1.1)

Definition A sequence Q:J is a function defined for all positive integers

An example is the sequence {nf(n + 1)}, in which Xn = nf(n + I), or

-I, I, -I, ... , ( -1 )n, ....